If the component under test crashes, then we can often be left in a state where the XMLRPC connection is still alive, but a response will never be given. But xmlrpclib unfortunetely has no timeout mechanism so the entire test system will just hang. We should have system in place that aborts calls (or even whole tests) that take too long.
lib/thintest/tlclient.py uses SIGALRM for timeout management. Could be an option.