That'd be interesting, but I cannot think of a efficient way to extract all the useful info from that dump.
hmmmm..... that could be a fun ML problem for someone to solve. Start with Collectibles, hand label the first 50 pages or so, see what it spits out, it's the kind of structured data it might work well on, especially with [auction] and [wts] labels already exisiting, then numeric bids and offers, almost always with a PK in the first post.
You could probably do a pretty good job just with a basic python script to be honest, especially if you added a "Report Mistake" button to the website, so you could effectively crowd source the labeling problem.