The issue is that each world should be a cluster of servers, and the application should be executed in parallel on multiple machines, preferably VMs, with a handful running each zone. If a zone starts getting congested, they spin up additional VMs and handle the load. It isn't rocket science.

If each world is truly a single server, then epic lulz is what we have.