bounded cognition in evolutionary game theory

Suppose you have a set of agents that, for behavioral and strategic reasons, all "cooperate" with each other, but would recognize if one of the other agents started to "defect"; cooperating would be an evolutionarily stable strategy in this context.  If the society gets larger you might expect there to be a point (Dunbar's number, for example) where the agents can't keep track of all of the other agents anymore; suppose, in fact, that we have 1400 agents, each of which is designed to keep track of 140 agents.  As long as all but a couple of agents continue to "cooperate", you're still fine; if the number creeps above 20 or so, then the information required to keep track of who has been cooperating and who has been defecting gets to 140 bits, and so one might suppose that would overwhelm the agents, and there would be a tipping point around there where cooperation would break down.