We already use forward-when-no-consumers to false.

In our particular scenario, we are using a cluster, and each client (durable subscriptions) maintains a primary and secondary connection to the cluster, only using one of those connections at a time.  The situation we have, is that when that client is disconnected for whatever reason, messages published for that client then round-robin between the to durable subscriptions on the cluster.  When the client reconnects, we can wait for the redistribution-delay to kick in to move all messages to the active connection, but then they are out-of-order from the way they were originally published, which is a problem.

We are essentially doing a poor mans HA, but without having to commit resources to a passive backup server.  What would be nice is to have the option of plugging in a load balancing policy on the server side, and in our case, when there are no active connections, just use a policy that picks one of the queue's and sends all the messages there, that way we can maintain ordering based on how the data was published.