I've been struggling with a multi-machine site for months... even with SP1 if I reduce the machines to 1, the services, both in Catalog and web REST endpoints retrieval are very fast. As soon as I add 2+ machines to that site (all VM's of the same specs as machine 1) the performance of the services drops dramatically, to the point of being unusable in any way. I'm attempting to configure a site that has 2 machines in one cluster for exposing map services to the public and one cluster with 5+ machines specifically for caching purposes.
From what I can tell, one of the big selling points of AGS 10.1 is scalability, flexibility. I'd like to be able to pragmatically be able to move machines from one cluster to the other based on load, caching regimens, etc... Can't do that if I have multiple sites.
Anyway, I sure hope 10.1 can iron this out. If not, it seems that 10.1 is the new Windows Vista. Please help prove that prediction wrong esri.