Respondents with high IQ seem more likely to have taken official tests, though; doesn't this overestimate LW's mean?

Any self-report will overestimate LW's mean, even if there is no disproportionality among test-takers. I've taken this into account with various assumed population means in the analysis of previous surveys, but there's fudging involved (if the average IQ of responders is 130, is it really sensible to expect non-responders have an average IQ of 100?).