After buildworld crashed I decided to try running Peter Holm's stress2 
test suite.

It produced some curiously-scrambled console output, such as
Mar  7 18:19:25 www kernel: 67c4m5d  trtiherd2  tpoi du se1 6n7o4n5- ptrreiseedn tt os cuhseed _nyoi

The striking feature is the intact date-timestamp and the mangled message
content. After much further output, including many lines of
(da0:umass-sim0:0:0:0): WRITE(10). CDB: 2a 00 00 26 35 e0 00 00 80 00 
the machine finally panic'd  and left a backtrace, which is recorded in 
console2. It has since rebooted with no obvious ill-effect.