The bimodal data linked to in my earlier post are, apparently, the result of a software bug. So disregard that post.
Suresh Venkatasubramanian has corrected data. It’s still very interesting, but not nearly as striking, and we now have a single more-or-less Bell-shaped curve.