I see that theory pop up from times to times, but it only really sounds plausible if we look at the IT sector in isolation. Here in Sweden around 85% of men
and 85% of women work in a profession that is gender segregated, and the national employment rate for both is practical identical.
Not only is 85% a really large part of the majority of everyone employed, and a higher number than in the US, but for any male dominated sector there is a equal large female dominated one. In addition, the trend has been for the last 50 years of ever increasing gender segregation.
If sexist atmosphere is the culprit then the numbers makes no sense at all. People like to throw in the gender paradox, but that is more of an observation rather than explanation. If we go by the data the cause must be very fundamental, exist in practically all work places, and have equal force at both women and men.
I personally ascribe the phenomenon to a pretty old theory from the 1970. People feel slightly more confident and secure in a decision when they mimic decisions of people they identify with. From the first moment someone choose a education path, to succeeding and failing with exams, to applying for jobs, to succeeding and failing in the job, each time the benefit of feeling more confident and secure apply a small bias. You thus get a leaky pipe, and the more equality in choice people have in every step the higher the probability is that you end up with a gender segregated work place by the time people are in the mid 40s or 50s. If we wanted to prevent this we would need to raise confidence of any minority (gender, race, wealth, background, anything that people identify with) to be identical to the observed effect. In my view this is the primary reason mentor programs actually work, while looking at atmosphere (sensitivity training comes to mind) is unlikely to change the outcome.