In a nutshell this is all because of math. Scheduling is NP complete and the only good algorithm we have is divide and conquer. Team size generally matches task

In a nutshell this is all because of math. Scheduling is NP complete and the only good algorithm we have is divide and conquer. Team size generally matches task size and tasks should be small enough so that all options can be effectively explored in order to find a (local) optimum solution. Larger teams generally mean larger problems which mean the exploration space can get too big to effectively explore before time runs out. On big teams with big problems you eventually have to ship something so you are forced to pick the option you are currently exploring which may or may not be good. The only time this really isn't true is when the solution is standardized so that the main blocker is just 'doing' the solution and not finding it. (An assembly line is probably a good example of this).