This turned out to be an issue with messages stuck in the queue. We had enabled graymail for a short while, and then turned it off. When we did this, it seems some messages got stranded and were attempting resends. We went into the queues and manually forwarded the messages to the original recipients. Once they were gone, the queues and counters cleared up. I also found KB articles for the web and email gateway appliances that indicated an smtp loop and cpu resource bug, and this issue seemed to fit those symptoms exactly, even though that issue was supposed to be resolved by upgrades (and this wasn't even the same product, but does seem to use the same engine).