Nobody seems to have a clue why the curve for females is less broad, and whether nature or nurture is the key issue.
One also has to ask the question is that curve real, or is an artefact of the tests and testing process? Despite the long history of intelligence tests, the ability of IQ tests to accurately measure 'G' (general intelligence) in an unbiased fashion is still an open research problem. It's easy to see that the tests themselves are as open to socialised gender biases as they are to cultural or educational biases. The latter is a well known problem with IQ tests and is one of the reasons it is very hard to create a good IQ test, as opposed to a good test of educational level in a particular culture.
Cultural biases in IQ testing can be very subtle. Consider a timed test of mental arithmetic. One would think that didn't have any cultural bias. However, different cultures count in different ways. To most of us here 4 * 20 = "?" has the obvious answer "eighty", but to a Frenchman it has the obvious answer "Quatre-vingts" or "four twenties". The francophone doesn't have to do any arithmetic, and so answers fractionally faster.
Extend that tiny cultural bias, that assumption that "all numbers are the same" to everything that one might use as a test of reasoning or thinking ability and it all adds up. I see no reason why social (or socialised) biases are any less likely to have an impact on test design than cultural ones.
Here's another example of a culturally biased question from a putative IQ test:
"An Englishman, Irishman and Scotsman walk into bar". This sentence is:
- The start of a short story
- The start of a joke
- The start of an essay on subcultural differences in the United Kingdom
- All of the above