Sure, if you gave the same test to a representative sample of LWers and to a representative sample of the general population, you could calibrate IQ scores across them. I still expect it to be less reliable than proctored IQ tests though, not because I'm worried about people lying about their scores, but because of a higher incidence of confounding factors such as distracting noises, internet connection failures, and even the presence of daylight from a nearby window.

http://h-m-g.com/projects/daylighting/publicity%20daylighting.htm

I suppose it might be interesting to include some IQ questions anyway, as it might still turn up some interesting results. We'd just have to keep the limitations in mind while analyzing the results.