The underscore bug sounds minor, but is not.
The problem with this bug is you can start with something that works just as you would expect, like "HTTP_PROXY is a CGI variable".
Now you go back and, edit the sentence: "HTTP_PROXY and REQUEST_METHOD are CGI variables". Suddenly, and in violation of Least Suprise, markdown spits out text that has a em tag between the two underscores.
So, in practice, you have to remember to always escape underscores, so you start with. "HTTP\_PROXY is a CGI variable" and never run into the problem.
Except now you have to keep this arbitrary rule in your head, and remember to do it all the time, and you end up with something that does not look like a plain text email; it's obviously marked up now. Which rather defeats the point of markdown.
The issue here is that Gruber seems to be satisfied with Markdown-the-language, and Atwood (and many others) aren't.
Obviously, Atwood could just create a new markup language derived from Markdown (like MultiMarkdown). But then he can't call it Markdown. He wants the best of both worlds, and he's not going to get it. And that's his problem, not ours.