Using some agent based simulations from a 2003 Nature paper, Chris Allen speculates that functional human groups actually have an expected size much smaller than the Dunbar number, said number being 150. I'm not sure about the validity of the conclusions, but was interested in the means of modeling and simulating group behavior and coordination. Seemed like a cool thing to hack up in a multi-agent simulator, or even just a good old scripting language.
Of course, some folks believe the Dunbar number is really wrong.