have you looked ad this article?
I haven't tried it yet, but this is something much closer to my idea of near caching. Basically you have two infinispan clusters - one is the near cache - it's embedded in application jvm (so it's really near) and it works in invalidation mode. The second cluster is called using RemoteCacheStore and it's in distribution mode.
Thank you for your reply.
Is the client cluister also using DIST mode?
my understanding is, that the client cluster is in INV (invalidation) mode. This way elements in the client cluster are near - you never need to get the element from another node in the cluster, but they don't become 'out of date' - if element is modified by one of the clients, it is removed from other clients' near cache.