jdporter helpfully sent me a note that last hour of cb was out of whack. I don't really know what the problem was (other than a lack of logging! I'll have to add that in as I find time), though the Windows-standard solution worked (restart the daemon, not the whole machine). Then I found this interesting tidbit. The time stamp associated with CB messages is in the future of what my clock says the current time is, by over 2 minutes. Corion pointed out the Tick Tock nodelet, and I noticed that the CB message time stamp is in the future of what PM says the current time is as well.
Maybe there's a server involved somewhere here where the ntp daemon isn't running/working, keeping the timestamps in sync?
Corion also seemed to suggest that the Tick Tock nodelet maybe should use select NOW() from dual instead of the localtime, but I'm not sure what type of extra hit the db would take on something like that. A working ntp daemon seems more prudent to me :-)
Note that the only effect this has on last hour of cb that I can be sure of is that anything from the future gets excluded in my SQL: "STAMP > ? AND STAMP <= ?" (so that messages that come in while I'm processing all this don't get included later).
While writing this, Athanasius also suggested in the CB that "Someone just posted a node “-3 seconds ago” !" - so the aesthetic effect is a bit bigger than just what I noticed. :-)