Further to JBAS-2128 we should also implement some sort of listener on the transaction timeout.

Given the stupid BMT SFSB semantics (this part of the spec is one my favourite anti-patterns
where a client can hog server resources.

sfsb.doBegin();
doLongRunningProcess();
sfsb.doCommit();

And even crash during the doLongRunningProcess().

We should allow the server to reclaim the resources from a transaction timeout more quickly.

Another option would be to allow a configuration that does a rollback() at transaction timeout.
But this would require some serious testing to make sure all JBoss code is checking transaction status correctly.

It would also confuse any non-j2ee compliant framework that registers transaction synchronizations
and then assumes their "thread local" is still valid from the timeout thread or they aren't threadsafe.
No names mentioned to protect the guilty