No idea if my experience will be of any help to you, but captchas aren't the only way to stymie a bot.
I was able to get one forum (not our mud's) that was getting a few hundred bot sign ups per week down to zero bot signups in a year, by replacing their captcha with a single static question, one whose answer couldn't be searched (some bots query Google and Bing for possible answers). I think my question asked the name of a specific menu item (or maybe it was a tab) in the site's layout, a question which most computer-literate people can easily answer. Since the menu and/or tab set that I was asking about was just a bunch of arranged links, the thinking was that most bots would be unable to make the cognitive connection between the arrangement of links and a "menu" (or "tab") as recognized by a human. So far, it's worked.
Of course, bear in mind that if the spam is targeted, it only takes a moment for a human to feed the correct answer to a bot.
I guess you just need a registration form with some questions (early mud codebases and their derivatives, names of the people who coded them, obscure segments of their licenses, math problems) that would be pretty hard for a bot to handle. If you still have the problem, then it's targeted spam.
P.S. Given the image captcha on the registration form, I think someone is being annoying on purpose. They either have a very intelligent bot and/or there is some kinda of free, internet-based captcha processor or they are expending a lot of effort to be obnoxious/get attention. Will ignoring them make it go away?
Bots have been breaking captchas for at least 7 years, to say nothing of all the ways they can dupe a random, relatively innocent person into doing the job for them. As bots improve, captchas have to evolve to stay ahead. I'd be wary of math problem "captchas" too, although the bots hitting MB aren't necessarily ones designed to deal with it.
Bots are usually unattended, and of course the bot itself doesn't care. Ignoring them usually doesn't change anything, but YMMV.