Yesterday morning, babylon4 started randomly crashing. And crashing, and crashing, and crashing, and crashing, and crashing, at intervals ranging from an hour or longer to as short as five minutes, with no visible connection to load or any specific activity on the machine. It wasn't logging anything nor creating any evident core dumps. I tried just about everything I could think of to diagnose the problem, without success, including running the machine through two full passes of memtest86+. (With zero errors.) I had a very strong suspicion it was a memory problem, except that I couldn't reproduce it with testing.
Finally, I happened to glance at the monitor in the rack at just the right instant, as the machine was rebooting from its previous crash, and crashed again. This time, I saw the kernel panic and dump core.
Having previously reorganized the rack so that I could get the cover off and check all the fans, I pulled babylon4 mostly out of the rack with the help of Eldest Daughter Formerly Known As Goose who now wishes to go by the handle Valkyrie (which suits her), and pulled and reseated all of the RAM. babylon4 has now been up 13 hours and 47 minutes without a problem, so it looks as though reseating solved the problem — fortunately, because I wasn't looking forward to having to replace all the RAM.
That said, I do still need to replace all the RAM at some point. Right now, the machine has 4GB of RAM (four 1GB DDR266 ECC DIMMs), but it actually has eight memory slots, and would run a lot better with 2GB modules in all eight. ZFS and SQL databases both like lots of RAM. There's no budget for it right now, though.