Nearly all MMORPGs use standard client-server models for networking, mainly because they are easy to implement and have well known properties (and hard limitations -- e.g., they do not scale). There probably have been some implementations that do ad hoc load balancing on clusters (for example, splitting up realms), but none of this is that interesting from an academic standpoint (mostly just \"hacks\").

A lot of the same ideas have been explored before in the context of FPS games. There probably is motivation to reevaluate this area since the latency assumptions in a MMORPG are drastically less stringint than in a FPS. If you just lookup \"games\" or something in a papers database (try citeseer), you will probably find stuff that is relevant (dead-reckoning, multicast groups, etc.). Projects off the top of my head are MazeMan (the first academic distributed architecture -- not client-server but p2p) and Mercury (http://www-2.cs.cmu.edu/~srini/Papers/publications/2002.netgames/) which implemented Quake using a publish-subscribe system over a Chord-like p2p network (Chord is the distributed hash table developed at MIT; but anyone involved in academia and is into systems or networking will know what a DHT is since they are all the rave now . Aside from that, I once read an article on gamestra about a clustered mmorpg implementation that some company was selling (and customers would just \"populate\" the engine to have a functioning mmorpg). Not sure if that was anything novel or not, but the PR release sounded interesting. There was some presentation about another MMORPG project over Pastry or something (another DHT) at Yale or someplace... it was shown at OSDI last fall... should be able to look it up.