The reactive flushing strategy flushes a transport using:

      TAO_ORB_Core *orb_core = transport->orb_core ();

      while (!transport->queue_is_empty ())
        {
          int result = orb_core->run (0, 1, ACE_TRY_ENV);
          ACE_TRY_CHECK;

          if (result == -1)
            return -1;
	}


unfortunately this loop has a race condition. The queue could be cleaned up by
another thread, before we can enter the event loop in orb_core->run(). 
Currently the problem only shows up when using AMI requests, for example,
$TAO_ROOT/tests/LongUpcalls.

A simple workaround is to use the Block_Flushing_Strategy.

The only solution that I can think of right now involves adding more complexity
to the Leader/Followers code, when the ORB flushes a transport it would need to
follow the same rules that "client" threads follow when they join the group.
But such a change is too complicated for me to take on at this point.