Empirically big groups do noticeably worse. I'm not 100% sure why, but my guess is that the problem is not with big groups per se, but what having a lot of cofounders says about the founders. I think when a startup has a lot of founders, it's often because the founders are afraid of starting a company, and thus recruit everyone they know to be in it with them.

Another theory I've heard is that a startup does as well as its worst founder, and the bigger the group, the more chance you have someone who's no good.

This is the famous tragedy of the commons problem where where the increase of members participating the in game (in an game theory sense), any individual is more likely to become a freeloader and hence less likely to put the optimal labor.

The fundamental problem comes from a form of information asymmetry where any individual is not able to assess the amount of labor placed by other members. If a project is run by 5 people and requires labor input of 5L, an optimal allocation of be L from each person. However, because each person does not know how much labor other people are placing into the project, they assume that others might put more than L, or the total labor from others are 5L-x where x<L. Hence the person in question only puts x amount of labor in the project. Alas, if everybody thinks in the same way, we only have 5x, which is far less than the required 5L. Unless all 5 people are together at all times, there are bound to be information asymmetry, and it is just much easier to manage with fewer people, in the case of 2-3.

I've experienced the second theory first hand. A bad partner is much worse than no partner at all.

Pretty much. Empirically, most startups with more than 3 founders collapse due to intra-founder strife.

(Obviously there are exceptions. I am a co-founder of one of those exceptions, since Justin.tv has 4 founders. But it's a good rule.)