Typically it's not possibly to load-balance JMS connection with a hardware load-balancer like is often done with e.g. HTTP. This is because JMS connections are stateful (and HTTP connections are not). When the client creates a connection there is an actual resource created on the server (i.e. the session) that is used by the client until the client closes the connection.
If your client is connecting through the load-balancer and the messages it send with this connection are only going to one of the servers, then that is likely the correct behavior. If the load-balancer tried to send messages to the other node it would get an exception because there would be no session on the server to accept the messages.
You may be right. But what makes me to worry is that WMQ receives messages in round robin fashion from the same load balancer. I wish to see fully functional fail over support with clusterning in HornetQ,