Linux

I think, your expectations are off for what a native integration would achieve. A kernel which has both a Linux API and a Windows API would be an insane maintenance effort. You'd naturally want the Windows APIs to simply be translated to the respective Linux API calls. This is what WINE does.
In theory, if it's directly integrated, you could do some dirtier stuff, i.e. call kernel-internal APIs, but you want to avoid that as much as possible, since those kernel-internal APIs are not nearly as stable as the public APIs.
It should also be said that writing kernel-level code is hard. You cannot ever crash, you cannot ever make mistakes when managing memory, you cannot allow yourself any vulnerabilities. Again, you want to avoid writing kernel-level code, if you can.

WINE has some additional ugly workarounds, like a virtualized filesystem. There's not terribly much you can do about that. Windows applications may simply expect certain folders to be in certain paths. You can't directly map that to a UNIX filesystem.

As far as I can tell, pretty much the only advantage of natively integrating it, would be that it's installed by default, which can be achieved in other ways (distros), and due to those ugly workarounds will not be popular at all. As much as I'm touting its horn right now, I do not want WINE on my system, unless I need it.

It's easy to be frustrated with WINE, because it does not handle all applications perfectly, and then think that the approach is just wrong. But yeah, no, some really smart folks came up with that approach. It's just insanely hard to get the exact (undocumented) behavior of the Windows kernel APIs correct, whether you do a mapping or implement them natively.