The challenge with applying this too broadly, I think is that it becomes a dual problem of objectively enough defining who fits in the marginalised class, and of defining a class that is big enough that you can try to accommodate it without harming other groups, marginalised or otherwise, in an unreasonable manner.

With respect to gender it's fairly simple. Already with respect to ethnicity it starts getting tricky to define objective rules.

It's also a tricky problem to apply at too small scale. E.g. in a small enough team or in a small enough niche, pure chance will end up with some small groups that are not diverse by pure chance even in an idealised setting with no biases. Trying to prevent that all the time could be potentially highly detrimental. Figuring out which imbalances and at which scales are by chance, and which are down to biases is a hard problem.