I'm using Tomcat 4.1 now outside of the JBoss JVM to make this work. Sadly, nobody responded to my inquiry :(
The 3.2 doc seems to insist this works inside the server VM but as far as I can tell that's not accurate.
If someone can chime in I'd be really grateful. I'd rather not have all this server indirection.
This is by design: jboss will use the target in the local JVM if it is available and, in this case, will not load-balance.
If you want two boxes, deploy EVERYTHING two times i.e. the app server AND the servlet server and put a load balancer in front of them, this way you will get more performances.