1. Message Identifiers/Primary Keys - if I recall correctly, IMAP doesn't have a way to definitively map a message and a unique identifier to the message in a way that is guaranteed to persist long-term.
2. Not exactly IMAP, but malformed messages. When I was working on this, I was using my GMail corpus, which goes back to about 2006 or so. Somewhere around 30-40,000 messages. Turns out it was a ridiculously good test case for parser robustness. Messages that claim to be ASCII-7bit, but have extended characters. Messages that claim to be UTF-8 but have invalid codepoints. Messages with malformed MIME headers. Messages with a text/plain section with garbage in it, while having a nicely formatted text/html section. Messages with embedded CSS that tries its hardest to blow yup on-screen display.
I think that through the duration of the project, I only had to submit 2 PRs to the upstream net::imap ruby library, but damn were there a lot of heuristics in place to try to successfully grab an email from IMAP and display it on screen.