Scaling problems that arise for multi-aget memory

Given a bunch of agents working together, how do you distribute the memeory, how do you update, what do you choose to remeber and how do you decide what to share and what not!