I don't think you'll break anything, but you'll lose the early notification of the failure.
i.e. in a lot of cases you'll be doing useless work which won't fail until it becomes time to commit/rollback.
That's cpu and resources that could be used by something else.
Perhaps you could make it an option in the -ds.xml instead?
The check was added to avoid sql operations leaking into the next transaction because of a race with the tx timeout.
This won't happen if the transaction is just MARKED rolled back only. The transaction hasn't completed yet.
See JBAS-5080, the race goes something like:
a) Thread 1: start running a query
b) Thread 2: timeout -> roll back transaction, which ends the local jdbc transaction
c) Thread 1: original query now reaches the real connection which is in a new local jdbc transaction as far as the jdbc driver is concerned
(b) hasn't happened if the JTA transactions is marked rolled back only