No idea anyone? Could that be a network/multicast problem? On the other hand: the servers recognize each other. And farming seems to work.
I'm guessing a network problem unrelated to multicast.
My interpretation is that your SLSB is deployed on both nodes. For an SLSB, there is no intracluster traffic related to it after it deploys. If it's deployed on all nodes, there is no intra-cluster HA-JNDI traffic related to clients looking it up. Those two facts eliminate a whole range of areas to explore.
Configure a jndi.properties on your client where you specify the host names of your servers; i.e. don't use HA-JNDI auto discovery
If you do that, there is no multicast at all involved in what you are doing. You're basically downloading a JNDI stub, doing lookups and making invocations that cause no intra-cluster activity on the server side. Not much difference from a non-clustered scenario. Only difference is the JNDI and EJB proxies will spread the calls over both nodes rather than targeting only one. If you find you have problems in this scenario, I would look for problems on each node standalone.
Thanks for your help! I finally figured it out! I uninstalled my software firewall (though it was definitely deactivated) and my virus scanner (also definitely deactivated). What can I say? One of them must have blocked something in between the two jboss instances. Now everything works perfectly. I can easyly see that the nodes are switched via round robin. Can you imagine the cause? This took me hours, and I am not smarter than before...
It's hard to guess. If my discussion of your environment was right, there should be no intra-cluster traffic after you've deployed the bean. I suspect whatever it was was interfering with communication between the client and one of the servers. But that's just a guess.