Motivation: I am trying to settle a design for my HornetQ messaging system and I want to avoid setups which would introduce lots of additional network roundtrips between cluster nodes and possible lockups. If there's a lot of overhead I would build a system ensuring connections to all nodes. Of course I would like to rely on the parameter and avoid the additional work.
You would transfer a message through the clustered queues without the need. that's the overhead. If you don't have a consumer on the other node is better that the message is not transferred at all.
Redistribution would eventually send the message back to where it's needed. But that's an unnecessary round trip.