I don't know what standard you have for "exceptionally more variable" but in a normal population you only need a 33% difference in standard deviation to get a r

I don't know what standard you have for "exceptionally more variable" but in a normal population you only need a 33% difference in standard deviation to get a ratio of 2x at +2stdev or almost 4x at +3stdev (here stdev is of the less-variable population). I think relative to the population at large it's reasonable to think a programming career is a +2stdev event in inclination and skill toward abstract systems (programmers are ballpark 5% of the working population). I don't mean to suggest that the male variability hypothesis is all or even most of the story, but to me a 33% difference in variability, while substantial, doesn't sound impossibly large — and it suggests male:female ratios that are in line with what we see in the industry.