As I understand it Agent Handlers are mini-epo servers. Computers talk to their agent handler and then the handler talks to the EPO server. With just repositories the agents will still contact the EPO server for policy updates and other policy information, but an agent handler will handle all communications including acting as a distributed repository. You will probably need a lot of systems, a very slow server, or very slow WAN links to really see the benefits. We've been running one EPO server and no repositories for 400+ systems across multiple sites around our county without issue, although I am adding repositories to help reduce bandwidth usage.