http://science.sciencemag.org/content/321/5888/494.summary https://www.jstor.org/stable/2889145?seq=1#page_scan_tab_con... https://www.ncbi.nlm.nih.gov/pubmed/21038941 http://www.ams.org/notices/201201/rtx120100010p.pdf
(Note: you need to read data tables, not titles/abstracts.)
Most of Alex's links support my claims about VAM - it's reliable, but not 100% perfect (some variance exists).
The remainder are easily available via google, I'd copy/paste sources but I'm not on my personal laptop at the moment.
Strangely, the original article didn't provide sources for it's claims. Why the isolated demand for rigor?
As to the first study (Hyde et al. 2008), your claim only holds for white 11th grade Minnesota students, and only in terms of standardized math scores. For Asian-Pacific Islander students, the M/F variance ratio (VR) at the 95%+ and 99%+ percentiles is 1.09 and 0.91 respectively.
As to the fourth study (Kane and Mertz 2012), they don't report VRs for the upper tails of the distribution; they focus exclusively on grade school children; and they report that the overall VR varies widely from country to country, with a mean of 1.08. So again, your claim does not follow from their results.
I haven't read the other two studies, but given the pattern so far, I don't think I'll need to.
Here's some discussion of that issue in a completely different context:
https://www.chrisstucchio.com/blog/2015/ab_testing_segments_...
https://www.chrisstucchio.com/pubs/slides/crunch_conf_2015/s...
Note that the article didn't do any statistical test on this number (with or without a degree-of-freedom correction). That's unsurprising, given the small sample size.
tl;dr; the subgroups are not large enough and the observed differences are not wide enough. Feel free to run a statistical test to prove me wrong.