The bug I referenced earlier was truly, truly stupid. I save blog entries in a serialized object file - one file per day, with each file being a collection of the post objects (with their associated comments, etc). When comments come in, the system looked up the existing entry, appended the new comment, and sent it off to the storage system for saving.
If you think about that, the problem is obvious. Say two people comment at about the same time. The first process gets started reading in the existing post, and is out on I/O. The second process does the same thing. They both append their comments to the post, and then they save. The save is serialized to prevent problems, but - in this case - the last one in "wins". That's why some people have seen a comment show up, and then disappear.
After I got a complaint about that, the problem came to me quickly - but I was in the car at the time. So this morning, I sat down to refactor the problem. The new code was pretty easy to slap in, and simplified the way comments get saved. So now the problem's gone - all that's left is the embarrassment over how stupid it was :)