Two broad reasons come to my mind.
A) Reality isn't just social construction. It has to be rooted in certain facts about individual biology and environmental demands. If biology was so insignificant that we could simply brush away existing bodily differences and they'd stop mattering, couldn't we then also simply convince people that they are of a certain sex or gender (for instance the one that popular belief assigns to them)? If one's perception of sexuality and identity were free floating entities "chosen" out of will and wish, then we could just coerce likewise people into believing they were of a certain gender (or by extension, persuade them that they aren't hungry and they'd just forgo starvation ).
That's the first reason. Environmental demands that the sexes were differently exposed to from an evolutionary perspective would have lead to systematic sex differences. How "relevant" they are is no doubt a matter of malleability (or plasticity) of nature, degree of systematicity in the different environments and other such complicated factors. (Not to mention that they'd be exceptions to this trend)
B) Secondly you need to understand thay social conventions are created to maximise happiness. Stereotypes, howevermuch of a bad word it may be today, are rom a completely neutral perspective only statistical predicitons or trends drawn from probabilistic data about social behaviour. And for all their fallouts, they do at the end of the day help increase efficiency and happiness.
Think of any group, say vegetarians. Now it might be good to offer them salad and government policies might be formulated to subsidize salads for them but it is also true that they'd be many exceptions who just despise all kind of salads. Is it unethical to stereotype them and aren't the policies unfair to those who are at odds with the majority group?
And that's it: a balance has to be maintained while targeting services or behaviour or products to groups based on the increase in happiness from this targeted/focused approach and the extent to which the members of that group actually share those characterisation (i.e. the reliability of that stereotype).
Now, giving scholarships to students' with high SAT scores seems justified (because the statistical trend of high scoring sat takers having proficient academic careers plays out in real life and the benefits of this "discrimination" outweigh any negative externalities).
But, disadvantaging people of a certain sex or race (women and blacks for example) depending on the differences in the average intelligence level might not seem such a good idea.
I'm not giving any conclusive comments on the justifiablity of these policies but the point I'm making is about the trade offs involved and how these are continuously contested themes.
So to summarize: gender matters, it cannot be changed completely out of willingness or political correctness, and it in many ways is a beneficial and productive social practice.