Bug in queued message subsystem?

[utzoo!decvax!duke!chico!harpo!npois!houxi!houxf!houxa!3255fah]

We are running UNIX 4.0 on a VAX as a base for a field application
system.  We have been observing a situation which results in a
corrupted message map.  By counting mallocs and mfrees (via hook)
we have verified that the queued message subsystem apparently will miss calls to
mfree at times, thus gradually "using up" the available space.

It appears (primarily through guesswork) that the failure to call
mfree may result from some variety of abnormal process termination,
possibly user stimulated.

Does this sound familiar to anyone out there ?  Any leads would be
appreciated ...... thanks in advance.