I haven't recovered at all either, after redesigning layout to feature content over ads and removing some cheesy SEO "tricks" like a script for auto-generating keyword variations for the titles of item pages - I don't know what I was thinking when I wrote that! ;-)
However, I think I know what is wrong and how to fix it. I have a three-level taxonomy. Top-level categories list mid-level categories, with a unique description and a link. These were not hit very hard. Bottom level subcategories list individual items, with a caption, image and link. These, for the most part, were not hit that hard either.
Mid-level categories, on the other hand, were hammered pretty much across the board. On a mid-level category page, I list the first two items in each subcategory, with the same caption, image and link that appears for these items on the subcategory page. There is also a link to the subcategory.
The upshot of this is that the mid-level category pages look like "tag" pages (although strictly speaking they're not). That is, they have content that is duplicated from the subcategory pages, no unique content except, in some cases, some fairly obvious spider-food at the bottom of the page which I should probably just delete.
This is obviously a big mistake, I should have seen it sooner. Now that I'm rethinking it, I realize that it's confusing to users and search engines alike and creates pages of questionable value at best. I'm changing the architecture so that the mid-level categories don't link to individual items at all, but rather just link to the subcategory page with a unique image and caption. That way the mid-level categories have unique content and links, and no duplicated content/links scraped from their subpages as they do now.
I'm also planning to add/upgrade lots of content in existing subcategories.
I think I've got it pinned down, I think it'll work this time... we'll see in another 4-6 weeks, apparently.