The failure actually didn't have anything to do with LargeMessage. The test was mixing LargeMessages with other types of messages.
The test was about the fix on https://jira.jboss.org/jira/browse/HORNETQ-304
If you received a MapMessage, and sent it immediately (without reading any property.. like the way JMSBridge would do), nothing would be read on the body, and you would see an exception.
To fix it, I'm setting HornetQMapMessage.invalid = false at the time of construction.
Invalidate was actually always true. So, I am also changing it to be set to false | true accordingly.
@Tim: If you could please take a look on the fix after I have committed it.