Usually, on my dual-core amd64 system (4.99.30), using LFS with, for
example, a logical segment size of 2-3MB, it will lock up (hard, e.g.
no keyboard LED response, cannot switch VTs) within 1-2 hours running
with a significant load average (1.00+).

I've been testing LFS fairly extensively to try to figure out
something about what's going on, as I'd really like to use it.  Last
evening, I did an installation of 4.99.30 (8/27 snapshot, GENERIC.MP
kernel) and left the machine running overnight building some packages
as well as a NetBSD release (build.sh -j4).  It has one LFS partition:
 a ~72GB root created with newfs_lfs' defaults (1MB segment size,
8192/1024 bsize/fsize).  It has yet to crash.
I also cranked up kern.maxvnodes a LOT, e.g. as if maxusers was ~512.
;)  The machine has not crashed.

Could anyone hazard a guess as to what the underlying modulating
factor might be here wrt: stability?  It would also be helpful if
others who've had problems with LFS on MP systems test this out.  If
it indeed improves stability that much for other users, IMHO we should
put something in CAVEATS/BUGS in newfs_lfs warning people about this
situation.

Thanks,

--Blair