This is a good theory. Where is the code is this hash table?
this is just a first idea. I think that i can take a look at the jboss-code in the next two weeks. And until then, i know only that, if the target entities of two or more 1:n-relations have the same pks, jboss has problems with cascade-delete.