We've seen infinite loops trying to fill same data over and over during session replication in a cluster. This data is a HashMap of org.jboss.portletbridge.RequestScopeManager$LRUMap. It looks like some concurrent access here to the RequestScopeManager could have thrown this HashMap.writeObject() call into an infinite loop to fill the bytes here(HashMaps aren't safe for concurrent access and can get stuck into infinite loops).
The RequestScopeManager has implemented a non thread safe LinkedHashMap. Please make this HashMap a thread safe one.