I don’t remember exactly the details of the test now but iirc the load test scripts were running for an hour or so. The load test scripts were implemented in Smack. Smack scripts were running from several “slave” machines (e.g. 10) and each on simulating a high number of virtual users.
Are they seeing memory leaks or just running out of memory? When playing with load tests it is easy to run out of memory and that is not necessarily a leak. For instance, the rate of incoming traffic is much higher than the rate of processing traffic. That will eventually fill up the MINA queue and consume all the server’s memory. For that reason I created the loadstats plugin where you can see some internals of the server to figure out what is going on.