I have played two sessions of Montsegur 1244. The first one, with 5 players, took a good 4:30, the second with only three players stayed well under 3:30 and included a significant break in the middle.
There are two main factors: Number of players…
The number of scenes is roughly proportional to the number of players, with some acts correcting the number of scenes towards 5 per act. This – in addition to the amount of general cross-talk going up with the number of players – makes the play time depend strongly on the number of players (but try to not reduce that number to 3: even though it's faster, it loses on the number of interactions.)
… and focus in framing scenes.
Beyond that, a large amount of the variance can be explained by the framing and playing of scenes.
In my first game, where none of us had experience, people found it hard to decide what scenes to frame and to keep scenes short; this was significantly faster after the break post-act 2, so we spend only about a third of the time on acts 3 and 4 and the epilogue and 2/3 on explanation, character choice, prologue and acts 1 and 2.
In my second game, I had brought a 3-minute sand timer to remind people of the fact that scenes should be short, but it proved to be unnecessary. A big point of the explanation here was to say “If your main character currently believes, try to frame a scene that might make her not believe; if she currently doesn't believe, what kind of scene might make her believe?” This made the tight focus of the scenes quite clear and helped decide what to frame.
I would consider a tight focus on scenes, and possibly on “show, don't tell” in the sense that scenes where the characters act are faster than scenes in which the characters talk to each other, a bigger factor on the play time than the number of players.