My intuition so far is that La Griffe found a convoluted way of regressing on a sigmoid, and the gain is coming from the part which looks like an exponential. I'm a little troubled that his stuff is so hard to reproduce sanely and that he doesn't compare against the exponential fit: the exponent is obvious, has a reasonable empirical justification. Granting that Dickerson published in 2006 and he wrote the smart fraction essay in 2002 he could at least have updated.

I emailed La Griffe via Steve Sailer in February 2013 with a link to this thread and a question about how his smart-fraction model works with the fresher IQ/nations data and compares to Dickerson's work. Sailer forwarded my email, but neither of us has had a reply since; he speculated that La Griffe may be having health issues.

In the absence of any defense by La Griffe, I think Dicker's exponential works better than La Griffe's fraction/sigmoid.