> Oh, just noticed something else via 'dmesg':
> 	sed[15050]: unimplemented software trap 0x29
> 
> Is this related to the 'illegal instruction' cores?

Yes, it is. It's a long-standing problem that I've been seeing for a
long time. It occurs sporadically and hits a random process.

The trap is classified as a "data error" by some sources, and as an
"internal processor error" by others. Some time ago, I found out that
you get it if you write to an undefined location in one of the alternate
address spaces. So, my current theory is that there's a code path leading
to the execution of some sun4/sun4c code (in a sun4m kernel).

-pk