I'm developing a solution that relies on Windows Server AppFabric's distributed cache features. Our current deployment scenario is to deploy use the same hosts for the cache cluster that our web apps and services run on.
This has lead to a few issues regarding memory availability (causing the cluster to throttle).
I can always throw more memory at the servers, but looking at memcached and other distributed cache cluster deployments, it seems the preferred strategy is to have dedicated cache hosts with limited server footprint, so that the memory can be gobbled up by the cache.
Server Core immediately jumps to mind, but I can't find any documentation supporting this strategy. There's a lot of "limited Managed Code support" in the documentation and it doesn't look hopeful.