Here's another stupid workaround:

/dev/sd0a	/		ffs	rw		1 1
/dev/sd0b	none		swap	sw		0 0
/dev/sd0b	/tmp		mfs	rw,-s=48000	0 0
/dev/sd0g	/usr		ffs	rw		1 2
/dev/sd1a	/mnt/1		ffs	rw		1 2
/dev/sd1b	none		swap	sw,-priority=1	0 0
/dev/sd2a	/var		ffs	rw		1 2
/dev/sd2b	none		swap	sw		0 0
kern		/kern		kernfs	rw		0 0
proc		/proc		procfs	rw		0 0

sd1b is a tiny, 2M partition created just for this purpose. If that line
is left out, init hangs after mounting sd1a and before setting ttys, just
as Henry B. Hotz describes. Curiously, it would work ok without it, if I
booted into single-user mode, did a "mount -a", and then "exit". 

Problems like this have been reported for some time; since 1.2.1 at least.
The only new twist is that it apparently has something to do with the
partitions, and therefore with our pseudo-disklabels. 

Why can't we have a disklabel in the NetBSD partition, the way the i386
port does? That system is very flexible. You can craft partitions outside
of the NetBSD partition, if you like, and it still coughs up a
pseudo-label if there is no real one. In the past, we haven't seemed to
need it as much as they do. DOS-like systems are limited to four
"physical" partitions, but we can have (seven or) eight. That's more than
enough for say, a 1G disk, but it starts to be restrictive with larger
drives.