In a clustered environment (i.e. JGroups "web" or "ejb" channels activated) the data contained in the session is stored either on a group of servers (distribution) or across all the servers in a cluster (replication). Distribution is the default clustering mechanism and it is governed by Infinispan which elects a number of owners to store the cache entries. As the cluster view varies so the distribution of cache entries can vary to ensure that data stored in the session is available.
By default all Stateful Session beans deployed in an ha/full-ha profile and web applications marked as "<distributable/>" achieve this behavior.