JBoss Cache is very much designed for the situation you describe. The problem with the hashmap approach is, as you say, the data can be changed. Maybe not frequent, but can happen. If it is changed, you need to reliably communicate that to all the servers in the cluster so they can update their hashmap.
Thanks a lot for your answer.
After reading more, it looks like it is a fairly easy solution to implement and will make my life easier later on.