ISPN-2643 made the java Hot Rod client always write to the main owner. ATM the client picks a random owner for reading though. This read-load-balancing doesn't really help, as assuming the data is evenly spread, the amount of reads would be distributed uniformly across the cluster. OTOH forcing the client to always read from the main owner would guarantee red-consistency for async replicated caches. Even so read consistency might still be a problem when a node crashed, but still these is a much stronger guarantee and makes async replication usable in many more scenarios.