ok, i further investigated the problem. it seems that it only happens, when the message is bigger than 8192 bytes and gets fragmented. this also happens with jboss code.
when i do 1500 updates on clustered entity beans inside one transactions, than the same long method call on the cluster happens for the invalidation framework.
so i guess it's not a problem of the code i'm using, but a problem of jgroups. any suggestions? maybe someone is using the invalidation framework for clustered entity beans and encountered the same problems??