Just getting some extra damage, and just missing, seem pretty boring. A lot of people like to spruce things up by having critical hits and natural-1 fumbles result in a roll on a special table, which causes Fun™ to happen. But the Dungeon Master’s Guide warns that these increases randomness, and notes that this is bad for the PCs. Why? What effects do such houserules have on the game? In what circumstances are they a particularly good or particularly bad idea?
The Dungeon Master’s Guide is right to warn you about such houserules. The game was not designed for them in mind, and the game’s math reacts poorly to their introduction. In general, some of the biggest problems are that critical and fumble tables
Increase swinginess, which disproportionately punishes PCs because they see many more rolls than does the average NPC
Increase the importance of individual rolls, which disproportionately punishes classes that roll more often – notably warriors who have lots of attacks
Happen at random, which means they often prevent interesting things from happening more than they introduce new interesting things
Know nothing about what’s going on in the game – the characters, the plot, the strategies, the tactics – which means they cannot tie into the game’s truly interesting things
The first two problems are mathematical, and the first one, in particular, is rather difficult to overcome. These problems exist without regard for table, playstyle, or other preferences; no matter your group’s ideas of what makes the game fun, these are still issues that are worth consideration. Over time, the disproportionate punishment that they yield makes it difficult to keep a campaign going – after enough rolls, there will eventually be a show-stopper.
The second two problems are more conceptual. Because the tables are “ignorant” – of timing, of the other things going on in the plot – they are very limited in what they can actually accomplish. Even for a well-designed table, that works out some system to account for the mathematical problems above, still cannot help but be rather boring – after all, they are unaware of any of the interesting things in the game. They often disrupt the game, and rarely add to it. And, most likely long before you roll enough to end the campaign, you are likely to have a critical or fumble that is just nonsensical and, well, silly. Silly is fine for many campaigns, of course, but a good joke is all about the timing. The tables, by definition, don’t have that.
Thus, critical and fumble tables are bad for any campaign that either wants to last for more than a few sessions, or wants to try to be serious at any point. They work against both of those goals. Thus, the only kinds of campaigns where these problems are minimized are those that are both short, and intended to be silly. In other words, critical and fumble tables deserve approximately the same considerations as the deck of many things.
The problems in detail
Increased swinginess – disproportionately punishes PCs
Critical and fumble tables are a major source of “swinginess” – low-probability, high-consequence risks. Swinginess is, flat-out, bad for the PCs; that’s a mathematical fact, noted by the DMG. The reason is because a given PC simply rolls more dice, or has more dice rolled against him or her, than does any NPC.
Generally speaking, “Team PC” and “Team NPC” (to include non-character challenges as well) will see more-or-less the same number of rolls – when a PC rolls, it’s almost-always against an NPC, and when an NPC rolls, it’s almost-always against a PC. This keeps the numbers similar. But, there are many more NPCs than there are PCs, and more significantly, only a fraction of the NPCs are “important” – any given roll by or against “Team NPC” is far, far more likely to be targeting “Goblin #893” than it is to target “the BBEG.” Meanwhile, every single roll by or against Team PC is by or against someone important. Everyone on Team PC is important.
So Team NPC and Team PC are equally-likely to see a truly disastrous critical or fumble. But Team NPC’s disaster will probably befall some random mook. Team PC’s disaster will, by definition, befall a PC. So from the PCs’ perspective, they trade a mook for a PC – that is never a good trade.
There really is no way to “fix” this issue with a typical table; it’s too one-size-(doesn’t)-fit-all. It simply leads to making it that much more difficult for the PCs to accomplish anything. And the more severe the consequences from fumbles and criticals are, the more PCs have to invest in protecting themselves from it – which is investment away from things that will actually improve the story, move the plot, or accomplish goals. These are all bad for the game.
Mathematically, a conceivable solution would be to weight critical and fumble chances to compensate for the mooks. I.e. PCs only fumble X% of the time, and NPCs only critical X% of the time, where X is the fraction of “Team NPC” that is “important.” As should be immediately obvious, there can be no one-size-fits-all solution to this weighting, and almost as obvious, figuring out the weighting for any given campaign would be really hard. For mooks and the BBEG, it’s obvious that the former are unimportant while the latter is very important, but for NPCs between these two extremes it’s rather hard to say which side they fall on.
Increased impact of each roll – disproportionately punishes warriors
Typically, any given roll determines just success or failure, nothing more. How important that roll is depends on how important whatever you’re succeeding on is. For a warrior with four attacks, any one attack isn’t that big a deal – he’s got another three. For a mage, that saving throw roll is far more important – it’s the only roll he gets for that turn, it’s all-or-nothing.
But most fumble tables (I say most because it’s conceivable to design one that doesn’t; I’ve never actually seen such a thing, however) treat every roll equally. That means that characters that roll more dice – which is supposed to be a good thing, it means you have more chances for partial success and are therefore more reliable – get hurt much more than characters who roll fewer dice. This is a problem, because as it turns out, in 3.5 the characters who roll more dice are already the weakest classes. Warriors are far weaker than mages, and this rule punishes them more.
Worse, it punishes warriors for being good at what they do. The more attacks you make, the more chances you get for a fumble – the more chances you get to have a disaster. This is absolutely inexcusable; getting better should not make you more prone to disastrously screwing it up.
There are ways to fix this. Eliminating the chance of fumbles from iterative attacks works well enough (except that a fumble on the first attack often prevents you from even trying the remaining attacks, which screws things up again). Or perhaps giving a character with four attacks only a 25% chance to fumble where he otherwise would.
Worse: fumbles only for attacks
Some fumble tables are only for attacks (ok, let’s be honest – most fumble tables). That means that only warriors are at risk of them. This is, obviously, massively skewed, and in the wrong direction. I would literally refuse to play anything but a spellcaster under such rules, and I think you’d have to be particularly masochistic to do so yourself.
Fumbles for spellcasting require, basically, that you “fumble” when someone nat-20s (and confirms?) their saving throw against a spell. These fumbles should, perhaps, be different from what warriors see, but something similar. And then the spellcaster “crits” when someone nat-1s (and confirms?) their saving throw. I don’t think these are good ideas. I don’t recommend them. But if warriors have to do it, the mages sure better be, because the mages have too many advantages as it is.
Related issue: goblin dice
Basically, a d20 has a lot of variance. Twenty different possibilities, each one equally-likely. That works well when you roll dice a whole lot (e.g. in combat) but really poorly when you roll one or two “super-critical” dice (e.g. most skill use). By increasing the importance of each die in combat, you introduce some of the goblin dice problem into combat, the one place where (previously) the goblin dice worked. I strongly recommend reading @Magician’s blog post there; it’s really useful and valuable food for thought on the subject of dice rolls and what the math may mean for your game.
Context ignorant – boring failure
Because fumbles and criticals are, by definition, random, they are completely ignorant of context. They don’t happen “because” it would improve the game, they don’t happen “when” it would be interesting, they just happen. At random. Whether that is good for the game, or not.
This means that, since you’re rolling lots of dice and having lots of dice rolled against you, and occasionally those dice are crucially important, sooner or later you’ll have a fumble or critical that really disrupts the game, usually by getting a PC killed. How bad or not PC death is depends heavily on gameplay preferences, but I think all can agree that a PC who dies holding off the horde so the others can escape, in a last-ditch desperate attempt to stop the BBEG’s evil ritual, or otherwise dies doing something awesome, is the best-case scenario. Meanwhile, I think everyone agrees that, even if not “bad” per se, a PC dying to a random critical off of a mook in some random speed-bump encounter is a bit of a disappointment. We like the heroic sacrifice; the random death doesn’t really add much to the game. Deaths in between, where the PC just pushed too hard, got too “greedy” to finish off a target and put him- or herself at risk, or failed to account for some tactic or ambush the enemy had planned, are OK and an accepted risk of playing.
Note the big difference between the “best” deaths, the “OK” deaths, and the “disappointing” deaths. The best deaths and the OK deaths involve the player making choices that make their death more likely, either by choosing to sacrifice themselves or by simply making a series of mistakes in a deadly situation. The disappointing death just “happens,” a freak accident. No meaningful choice on the player’s part contributed to the death, excepting, perhaps, the choice to be an adventurer (read: play) in the first place. Critical and fumble tables dramatically increase the chances of “disappointing” PC deaths. And at the extreme, critical and fumble tables can turn Dungeons & Dragons into War Games.
Ultimately, we accept some risk of “disappointing” death in order to add some uncertainty to the game; knowing everything that’s going to happen before you do it takes the fun out of the game. But it’s a risk we have to consciously consider – the designers certainly had it strongly in mind when making the game. That’s why, in the DMG, they recommend against critical and fumble tables. Such tables upset the balance between having enough uncertainty to be fun, and keeping the risk of “disappointing” death low.
Finally, it’s worth mentioning a little more why “disappointing” death is disappointing; I’ve just asserted that, appealing to your emotional reaction to the idea. I imagine that most people are on board with that, but it deserves a little more discussion. Ultimately, we’re crafting a story when we play an RPG. The Player’s Handbook and Dungeon Master’s Guide go on at length about this. Random, “disappointing” deaths don’t add to the story; they halt it. They force the player to stop playing, waiting until either a new character can be introduced or the old character can be resurrected. The other players are forced to stop with the plot and either meet someone new, or else go off to resurrect their fallen comrade. If the story of how the PC died is interesting enough, this is fine – a break in the action, especially after a dramatic death, is appropriate. But if they were in the middle of trying to accomplish something, and the PC died randomly, that just disrupts the game.
All of this means that critical and fumble tables actually contribute to the problem of “boring failure” – particularly the fumbles. The whole point of a fumble table is to have something “interesting” happen instead of “just” an auto-miss – which is, certainly boring failure. But the fumbles aren’t actually interesting, they mostly just lead to more boring failure. Even if you don’t die, the fumble prevents you from doing what you were going to do. That’s boring failure. And fumbles are basically required to be boring failure, because of the following:
Limited benefit – don’t tie into any of the interesting parts of the game
What are the interesting things in the game? The plot, hopefully, and the characters, both PC and NPC. There are, hopefully, great heroes, dastardly villains, and an impressive scheme going on. And then there are the more mechanical bits – the strategies employed via classes, feats, and items, and the tactics in a given fight, the maneuvers used and spells cast.
Critical and fumble tables don’t interact with any of those things.
They’re random and drawn up ahead of time. They are, again, one-size-(doesn’t)-fits-all. By definition, they cannot be specific to the characters involved or the plot that’s going on. They don’t react to the particular strategies or tactics that are going on. They just are. They’ll be the same in every game.
The characters can react to them; that much is true. That can be interesting – sometimes. But they don’t just happen when they’re interesting. They just happen. At random. And that really limits their effectiveness as plot points – because they just don’t tie into the plot.
And the rest of the time, they are doomed to being boring failure because they just don’t have access to anything more interesting to interact with.
But I Don’t Play the Way You Do!
Of course; there are infinitely-many different ways to play an RPG, and we all have our own preferences. These get broadly lumped into categories, though it’s worth bearing in mind that truly, no two people are wholly alike.
That said, I maintain that the problems with critical and fumble tables are so severe, and their benefits are so minimal, that they are bad for your game no matter what you play preferences are. The only exception, already noted, is a short-running, silly game – for when you want to use D&D rules to play Paranoia. That’s a fine thing to do (though Paranoia itself has better rules for it), but for all other games, introducing critical and fumble tables will make that game worse, and, if they have already been introduced, removing them will improve it.
I will use the same categories as mxyzplk for this exercise; they’re not entirely ideal in my mind, but it makes it easier to directly compare the two answers.
Mxyzplk paints a caricature here that I find obnoxious, and I have no idea what it has to do with “lawful” behavior, but nonetheless, if you are playing a game where the PCs are supposed to kill lots of foes and are not supposed to “lose,” then yes, we agree that critical and fumble tables are bad here.
Who I believe are being described here, and are thus being not-so-subtly insulted, are those who play the game more-or-less by the rules as written, and otherwise have a fairly legalistic relationship with the rules. People who expect the rules themselves to “work out” as they stand and judge rules by how well they hold up under scrutiny. Well, again, critical and fumble tables don’t stand up to much scrutiny at all They are clearly imbalanced, both between PCs vs. NPCs and between different classes. This, of course, has nothing to do with “winning” or “losing” but simply how effective these rules are at improving the game; they aren’t.
I think this group is, again, being caricatured unfairly – it could describe a much wider group than simply those playing with The Chosen One™. This describes any group whose idea of a fun D&D game is one in which the characters have goals, work to accomplish those goals, and hopefully, through much adversity, accomplish them. That could include The Chosen One™ but it also includes a far broader set of games. In fact, at a guess, most of them. Certainly in my experience.
Anyway, due to their random, context-and-plot-ignorant characteristics, critical and fumble tables add very little here. They do add disruptions that prevent things from happening. Since the goal here is for certain things to happen, that’s a bad thing.
This group is contrasted with the “lawful” narrativists, in that they don’t come to the game with goals, but rather let those come out in play. I think that’s a false dichotomy, in that A. pretty much everyone has goals; any fleshed-out character should have something even if it’s just to bring in the harvest or get another pint, and B. pretty much every game should have events happening that change things, that add new goals or make older goals less important, unimportant, or even undesirable, by virtue of things happening in the game. I am extremely dubious of the claim that any game lacks either.
That said, for a preference, we could call the distinction between the two groups that the “lawful” side writes big, important, well-established, well-defined goals into their backstories, and that the player’s goal for the game is for the character’s goal(s) to be accomplished, while on the “chaotic” side you have less important and/or less defined goals, that can be thrown to the way-side as the game progresses and more important things come up.
I still say most characters are a mix of the two.
Anyway, if you want things to react to, things to come up that you can decide how your character will react... critical and fumble tables add very little to react to. They’re random and arbitrary. They have to be extremely minimal if they’re to be playable at all, but that means they aren’t really worth changing your world-view around.
But say something big does happen – that basically means someone died. The “chaotic narrativist” gets to react to his ally falling through the cruel whims of fate, hooray? Maybe for him, but it might suck for the player of the character who died. But we’ll stipulate that his player is also interested in how the party reacts. This is good, right? This is why we should have critical and fumble tables?
Wrong. Because they’re random and arbitrary. Because they don’t have timing, they don’t have awareness. They simply happen. And that means, after you’ve had this great moment where the character who hadn’t had real investment now wants to carry on his ally’s quest – or whatever his reaction to the death was – it happens again. And the second time just doesn’t mean as much. It waters down the effect of both.
D&D is a game overseen by a real, thinking, critical human being. The DM can, and should, push the PCs in ways that will intrigue the players, get their investment. The DM can, an d should, make calls that will give characters something to react to, and grow from. That is what the DM is for. The dice cannot replace the DM, and they do a really poor job of it. Considering all of the problems that exist with critical and fumble tables, the poor job that the dice do, the little that they have to offer, are just not worth the risks involved. The DM can do a better job without them.
Ah, here is our Paranoia game. At least, I assume that’s what mxyzplk is referring to, since he says they’ll appreciate critical and fumble tables. In a game of Paranoia (or a game of D&D emulating Paranoia), go ahead; that is a use-case for critical and fumble tables. I still say Paranoia, itself, is better for a Paranoia-style game.
But then he calls this game-style “gritty” and “hard mode.” Well, Paranoia’s not really that. Ah, so everything-stacked-against-the-PCs? Well, that can be accomplished far better without critical and fumble tables. Why? In this case, because of the inverse: NPCs fumbling, PCs critting and insta-killing, take away from the challenge. Per my first “problem” above, this is less likely than the PCs doing so, but it’s still a risk that’s just not necessary.
It’s not necessary because there are better ways of accomplishing a challenge. Harder, more intelligent enemies. Enemies in terrain and situations highly advantageous to them. And so on. “But wait! Why not have those things and critical and fumble tables?” Because you cannot have a no-holds-barred 3.5 challenge mode. There must be limits imposed by the DM; a DM going all-out can kill the PCs any number of ways with absolutely zero chance for the PCs to do anything about it. The DM can bring in demon lords, angelic forces, and actual gods. The DM can bring in Pun-pun, or any number of other absurd Theoretical Optimization tricks. For that matter, so can the PCs – they have to be limited somehow as well, or there is no challenge.
So you can’t just throw everything at the PCs. If you do, they’ll either die, or have to come to the table that, by strict RAW, can’t lose (and yes, there are a few of those). Critical and fumble tables are low-value distractions that detract from the PCs ability to face more interesting, more dangerous enemies. And risk eliminating an interesting challenge as collateral damage, robbing the players of a real challenge.
Again, critical and fumble tables are random. They are completely ignorant of what is going on in the game. And I have never seen one that didn’t have results that would be completely nonsensical in certain situations. There is a reason I said that a game with critical and fumble tables has to be silly – critical and fumble tables are.
How many fumble tables have you seen that includes you attacking yourself? Most do, that I have seen. Have you ever considered how you do that? Your attack damage involves a carefully-aimed swing, with all your might behind it. Body mechanics says you simply cannot get that kind of leverage in some swing that’s supposed to come back and hit you. Or are you accidentally flipping your sword around and stabbing it into your gut, seppuku-style?
Or how about the common fumble theme of attacking an ally? There are plenty of cases where that’s entirely possible, even likely such that a skilled warrior would have to be very careful, or avoid the blow altogether. Definitely. Except that your roll probably won’t come up “attack an ally” when that would actually be a risk. It will probably come up “you are shaken for one round” or something. Instead, you’ll get “attack an ally” when your ally is behind you and to the left, relative to where you were swinging at an enemy. How do you describe what you were doing where that was a conceivable risk?
These are just illustrative. Critical and fumble tables fail miserably for simulation, because they don’t simulate the current circumstances, they apply the same risks to every situation, equally. They might yield an appropriate result, but more likely they won’t, and sooner or later they’ll yield something truly immersion-breaking and nonsensical.
A simulationist approach might include special fumbles or special criticals, but not off a table – it should instead be determined by the DM and/or the group, whatever is realistic for the given situation. Then you avoid breaking the immersion by having a stupid roll come up.
Averages and Odds
The problem of fumble tables, or increased critical effects is about averages and consequences.
For any given individual NPC or monster they make maybe a dozen or two dozen rolls in the time they appear in the game. Player characters, on the other hand, persist through several adventures, make way more rolls over time, and therefore play the odds more, and the more swingy the odds, and the higher the consequences, the more likely it is to disrupt long term play.
3.5 attempted to keep criticals but reduce their swinginess with the second roll to confirm the critical hit, then making the bonus damage a second roll. All of this is attempts to reduce criticals and reduce the amount of damage spiking that can happen when it does happen.
How it affects the actual people playing
The monsters or NPCS are all replaceable - the cost to them rolling a fumble or someone scoring a critical on them, and removing them from play is not high - it's a frequent part of play and the default expectation for such characters.
When PCs die, they may die permanently, or it may be some time before they can be brought back into play via resurrection etc. During that time, you have a player who is sitting there doing nothing until either their character is raised or they can have a new character introduced into play.
The big difference between these two situations is that the former doesn't leave anyone at the table doing nothing, while the latter does. And when you have to take time to coordinate to get all your players together at the same time to play - whether face to face or via online video chat, it is a larger overhead of time and direct impact on the fun real people are having.
Really old styles of play and randomness
Much older styles of D&D often had each player controlling several characters. This meant a lot of the swingy random issues - random attributes, trying to meet class requirements, imbalances between class types, didn't matter because each player ended up with some good characters and some weak characters by the laws of averages over several characters.
It also meant that the high lethality of the game wasn't a play-stopper either, because if one of your characters died, you still had several others you could call upon, or at least, one of your other friends could just give you one of their roster of characters as well.
Once you knock it down to players having one character to one player, character survivability becomes a bigger issue, since you start losing playtime to death, and swinginess becomes significant problem.
What are we reaching for?
Usually critical/fumble tables are added for one of two reasons:
1) Underdog's Chances
Criticals and Fumbles can let weak creatures take out stronger creatures easier. This is a shortcut towards dealing with the usual problem of hit point buffers and so on. This is a big issue in 3.5, especially as you go higher in level and the differences in math as far as attack bonuses, saves, hit points, etc. increases.
2) Chaos in combat
The unpredictability you mention as "fun". 3.5 in particular leans away from chaos in combat - battlefield movement is usually slow and predictable under the Attacks of Opportunity rules, characters are usually optimized towards doing 1 or 2 things, over and over in a fight, and so on. Randomized events does add chaos, it just happens that it ends up costing the player characters more painfully than not over time.
In both cases, these struggle with key points of how 3.5 D&D works, and I think altering them requires significantly deep rules hacks to resolve, definitely more than simply adding fumbles or criticals to the situation.
Some games avoid this issue by making swinginess in combat have some plot armor for player characters specifically. For example, the old Warhammer Fantasy Roleplay game had exploding dice damage, which is pretty swingy, but at the same time, gave player characters points they could spend to instantly ignore an attack and survive it.
Games like Burning Wheel use things like having players pick tactical choices and reveal simultaneously to produce chaotic fight scenes without necessarily going towards more swingy math.
First of all, let's check what the 3.5e DMG says. It provides critical misses as an optional rule.
VARIANT: CRITICAL MISSES (FUMBLES)
If you want to model a chance that in combat a character could fumble his weapon, then when a player rolls a 1 on his attack roll, have him make a DC 10 Dexterity check. If he fails, his character fumbles. You need to decided what it means to fumble, but in general, that character should lose a turn of activity as he regains his balance, picks up a dropped weapon, clears his head, steadies himself, or whatever.
Fumbles are not appropriate to all games. They can add excitement or interest to combat, but they can also detract from the fun. They certainly add more randomness to combat. Add this variant rule only after careful consideration.
It does not "recommend against them" - if it did, it would not have provided the rule in the first place. It recognizes certain games will benefit from them and others will not and encourages you to use critical thinking as a result. Let's unpack why.
- Randomness - Critical misses mean increased risk to the PCs in combat. They also increase the unpredictability of outcomes for the GM.
- Disparity - With any source of increased randomness, it will affect PCs a lot over the course of their lives, as they make many rolls. It will affect PCs that make a lot of rolls (e.g. fighters) over those that typically don't (e.g. wizards).
- Randomness - Critical misses mean increased risk to the PCs in combat. They also increase the unpredictability of outcomes for the GM.
- Diversity - Adds variety (which can add fun/challenge) in combat; instead of pure hits and misses, you can have more colorful setbacks.
The real effect on your game comes from two factors: the type of game and the frequency and magnitude of critical failures.
Lawful Gamist - Many people who play D&D and its variants are doing so because it provides for power fantasy. They want to engage in many combats and kill lots of foes, ideally without the fun being disrupted by "losing" (death, fumbles, significant level loss, etc.). For this kind of game, fumble rules are obviously bad.
Lawful Narrativist - Other gamers play for the story, but in the sense that they want to impose a story onto the game. (This is usually a GM concern.) If a campaign is based around a character who is The Chosen One (tm) then it's catastrophic to the plot if he or she dies to a random owlbear encounter. In a game like this, fumbles are bad - and really any other source of randomization is bad, so you might expect GM fudging and other frequently-discussed activities of that sort to mitigate all randomness.
Chaotic Narrativist - Some "story" gamers instead appreciate the story that is emergent from the characters' activities rather than wanting to impose one on them. For these games, interesting unscripted additions add flair and realism to the story. Narrativists are less "attached" to the character's fate than the overall story, you can see games like Fiasco that encourage you to put characters through all kinds of bad stuff, and people find it enjoyable. There's a gamer in my current group who just lives to see bad things happen to his character - he gets enjoyment from that.
Chaotic Gamist - Aka "hard mode" or "gritty." Some games want any combat to be risky. The high disparity between levels isn't a virtue but a problem, and to implement more even play (like GURPS and other systems), adding anything that means risk for you when you pick up your sword, even if you're 16th level and it's "just a bunch of hobgoblins," is desirable. Variants like E6 try to accomplish a similar goal. Or old school games like DCC, where you start with a whole bunch of PCs per player with the express goal of running through the meat grinder.
Lawful or Chaotic Simulationist - From the sim point of view, to the degree that fumbles help emulate real combat situations they are good and to the degree that they don't are bad - I don't think there's two different schools of thought here. So a fumble system that is based on real world probabilities and combat results is good (see "Keeping Their Heads Down" in this e-zine as an example of an article implementing friction and quoting stats from James Dunnigan's How To Make War).
As you will note from my chosen metaphor of lawful and chaotic, there are many gamers who fit into the neutral middle ground of these extremes. A neutral gamist might say "well there's a lot of things in the game that affect randomness, it's really the disparity against fighters that's the big deal to me." A neutral narrativist might say "well, some spice is good, I've given my PCs some Hero Points already as a narrativist currency so they can get out of a really pointless downturn." The main point is that there's not just one approach to gaming, so there is no way to make a fundamentalist declaration about whether this or any other mechanic is absolutely good or bad.
Frequency and Magnitude
Obviously the impact of fumbles can be affected by the conditions under which they occur and the kind of effects they have. If you make people roll to do anything, are doing fumbles on a natural 1, and effects include "you decapitate yourself," then you have introduced a huge amount of randomness. If you have reduced rolling, add additional gates to the fumble (the original rule's DC10 DEX check, a reroll to hit requiring a miss, etc.) and effects are "you are dazed for a round," then you have introduced a small amount of randomness.
This can be varied based on the mood of the game. If it's a Planescape planes-hopping game, then perhaps there are fumbles on Chaotic planes but not even crits on Lawful ones. Or impact on the mechanic from other factors (blessed/hallowed areas vs cursed/unhallowed ones).
Disparity is the real thing to watch out for here - sadly the base D&D system has physical combat and skills always require rolls while magic and movement do not... This can be tuned by understanding the problem (apply it to SR-overcoming rolls, limit to 1/round on physical combat folks, etc.).
Real Life Experience
Here on RPG.SE we like random theory to be backed up by experience. So here's how I've used fumbles and their effect on the games they were in.
The Pirate's Curse
In one game (Pathfinder, but the difference is irrelevant for these purposes), the normal rule set had no fumbles. I introduced a character known as Jaren the Jinx, a cursed pirate whose help the party needed. When he was around, fumbles started happening - both on natural 1's but even on failures to confirm critical hits. This was a significant problem for the pirate crew! They managed to use magic to mitigate the effect down to just fumbles on natural 1's but not past that; naturally everyone hated him as "cursed" and a PC eventually slew him - but got the curse in the process (everyone knows if you kill a cursed sailor you get his curse). He finally got it lifted by the witch who imposed it in the first place. This use of fumble rules was a direct support to the themes of the story and the risk it added was "undesirable" from an in-character point of view but extremely desirable and effective from an overall game point of view.
The Night Below
In another very long term D&D game, it was a deliberately low-power, low-combat, low-magic game. It was a D&D/Call of Cthulhu crossover game and needed to capture some of the feel of that genre (woot Cthulhu d20!). We used fumble rules throughout because it was appropriate for there to be risk anytime one resorted to physical combat. Mages rolled Spellcraft when they cast spells - instead of being able to perfectly place that fireball right on the "squares you want it," I used Spellcraft and the grenade-like missile bounce table to make magic more random and threatening. (Which eliminated the "roll disparity." Again, a game where the increased randomness and risk were desirable. It wasn't a game built around the power gaming principle and that' wasn't the expectation of the players in that campaign.
Fumbles don't fit every game - like the DMG says. They do fit some games. I'm not even going to say "some gaming groups," because why do you need to be an obsessive fetishist and always play the same game the same way time after time? I have run and played in many fumble-free campaigns, and also run and played in many with fumbles, and never was 'the fumble rules' the reason why I enjoyed or did not enjoy a given campaign. And in the end, GMing is not served by absolutist "YES!" or "NO!"s - it's a subtle art, where different amounts of all the available ingredients are appropriate for different games, campaigns, players, and situations.