The idea has potential, but the real problem isn't deciding how many classes to have, there are multiple problems.

For one, this method may be a lot of extra work to wind up with the same number of, or potentially even more edge cases. And chances are good you'll wind up with virtually the same results for everything but the edge cases.

Deciding what criteria to use is also non-trivial. There are a number of intangibles that may matter. "Women's Champion" isn't the same as "Class B Champion." Saying the "best female" conveys something much different than "best in statistical category F."

It does now because we've never split things up this way, but it might not be so weird once we get use to it.

On the other hand, I see a bigger problem. If we are using this technique for the 100 m dash (for example) then we're going to basically split people up into "best category", "second best category", etc... and we'll end up comparing between categories anyway since there is an objective measure to use (the clock).

That bigger problem is exactly the one I was talking about. Competition sorts people by performance already. Adding another new layer of sorting based on computer analysis of last year's performance doesn't really gain you much.

Virtually all sports rely on some level of arbitrary meta-organization. While you could write essays about why that is, it boils down to one: It's interesting that way. It's not just gender, either. Playing fields across most sports are divided by geography, age, affiliation or nationality, or simply arbitrary divisions like in the MLB. So long as the boundaries are recognized by everyone there isn't a problem.

I don't think that's the only reason sports are divided up. It's also because you need people of nearly equal skill before the competition actually is competitive and not just one player trouncing everyone else. Statistical methods would ensure that this happens, but may lose the other sort of divisions.