With a warm cache (with no eviction) we should never need to hit the CacheLoader again.
Was reported - specifically with DIST on multiple nodes - that occasionall IO is triggered; we should add tests to verify this doesn't happen (or to understand why it should)
I'm suspecting demand of null valued entries.

mailing list thread: http://lists.jboss.org/pipermail/infinispan-dev/2013-March/012261.html