I think this problem is due to the switch putting trap frames into the processing queue instead of the CPU port's output queue. This might be related to the pause frame nonsense I observed last week. Very interested to know if a 2.4 exhibits the problem.