Exactly. The simple arithmetic is correct, but the applied maths and logic is a disaster.
To work through your 'Rapist Bob' example, let's imagine a company of 99 women plus rapist Bob, and with Leonardo's bizarre 2-person-single-occupancy loos.
If men commit 98% of sexual assults, then at first look you'd expect 2 offences (rounding up) by women for each one by 1 by Rapist Bob.
But what's the pattern of offending? Women who commit sexual offences don't generally attack other adult women - so the chances of a workplace assault by a woman are much, much lower than that headline 2%.
If you're in the 2-person loo with 1 other person, and that person is a woman, the probability is that it's one of the 'safe' women. And if it's not, you have a much higher chance of fighting her off than you have of fighting off a man. So the fear is considerably lower, as well as the actual.risk.
And if we go back to Leonardo rather than Rapist Bob's company, the big error of logic is that the comparison of numbers is wrong. If you let any men into the women's loos you let all men into the women's loos - 80% of employees, not 0.5%.