It depends on what you use.
In case of EJB clients the JBoss API will do the work for you.
In case of http clients you need a balancer in front of your JBoss and there are many ways to do that.
I'll try to clarify my problem: particular problem is not how load balancing is done(it's documented), but solving "single point of failure" for load BALANCERS themselves. How can second balancer service a client when first balancer's service dies?
I'm not sure but AFAIR I've seen this is possible by switching the virtual IP or DNS by an aditional script or a server-cluster solution