Steve,
You may be running into a situation where output / input packets are being dropped at the interface after a fault condition. Sort of a wedge. Do you know what the input / output queue counters read after this fault condition? Also, is it possible for you to debug frame LMI when this occurs to see if the LMIs are being properly exchanged with the right sequence numbers? You may be hitting a caveat, and really like you discovered, the serial restart-delay is not appropriate here.
Victor