Well, setting NKMEMPAGES to 48000 caused the crash to
happen much later in the build.   Note that unfortunately
there is no core dump or anything, however as with the
uniprocessor crash, the number of short files found by
fsck in /usr/obj.i386 increases as NKMEMPAGES is set higher.

(Seems consistent with options SOFTDEP eating all kmem?)

All told, though, so far the multiprocessor kernel is
behaving not much worse than the uniprocessor one under
the same stresses.   I'm impressed!

        Sean.