Ah yeah, I check CDF and similar metrics against the user stats data all the time.  Part of the issue is that many many accounts have never submitted a block, and most are too slow to reasonably have been expected to.  The ones that have submitted blocks, and have mined long enough to have useful stats, are all within reasonable ranges.  I've correlated using usernames, along with non-public metrics... but the network difficulty is just too high to get meaningful results.  The fact is that most miners just won't ever submit a block anyway.  With 85% of users being under 4Th/sec, that's something like 3 years on average for those users to find a block, at the top of the 85% list.  There's also nothing preventing anyone from changing usernames and whatnot.  So for 85% of the users (~50% of the pool) most of them will never solve a block.  Only a small portion of them will.

I've checked accounts that haven't solved blocks by checking the difficulty achieved by all of their shares in the database against their expected amounts, and haven't found anything too crazy in that data either... although that data is much shorter-term than it used to be since I don't have unlimited space for storing full share submissions indefinitely.  Hardware that screwed up randomly or purposely wouldn't necessarily be noticeable this way anyway.

Edit: I actually misread your post and just realized you're talking about per-share stats.  I've done something similar, but I don't think this would help with potential winning share withholding detection for users who haven't submitted any blocks... or am I missing something there?