[eluser]bretticus[/eluser]
If your discussion is threaded, you can send the email with the ID for the current thread. That way you only have to compare the last thread if your email ONLY contains the last thread in the initial notification. If you're going to include them all, you could (and this is off the top of my head) strip all alphanumeric characters from the post initially and take a hash and store that in the database. Next, you'd have to do the same stripping and loop through the returned email looking for expected patterns and reducing the text (ouch!) Even then, I'm not sure that's fool-proof. Perhaps just pay attention to the several systems out there that already do this and see if perhaps, there's some patterns to detect quoted text, etc.
Quite a problem. Good luck!