So I have a funny wild theory...remember back when the Gawker database was compromised? And LinkedIn forced a password reset for users who (according to what I read) used email addresses that matched the Gawker leak?
What if they also (or actually) compared password hashes from their database to the ones released in the Gawker breach? In that case, they likely wouldn't have pulled data straight from the database but actually might have pulled passes from the db, output to text files, cut the text files up to parcel out for processing via Hadoop or something? And somehow one of those text files got loose somehow...or someone MiTMed the actual process (I'd vote for a floating text file just because it's been so long; the Gawker breach was in December 2010).