I just had a look at it. I picked a file at random - auth-passwd.c and looked at the Microsoft version versus the equivalent from the portable OpenSSH project. The Microsoft version is nearly twice as long (378 lines versus 216), with the addition of huge #ifdef hacks, converting back and forth between UTF8 and UTF16, and all sorts of other horrors.
If I look at the altered source in all the files as a whole, there are 363 #ifdef or #ifndef sections related to what has been hacked in for Windows support. Other platforms (e.g. Linux, Apple, AIX) manage with just a couple of special cases.
I'm not sure the OpenSSH developers would want that sort of thing hacked into their upstream source base. If it was me, I would certainly turn it down until the MS developers figure out how to separate that sort of stuff out in a minimally intrusive way.
I'm really hoping they can get this working, as getting the server daemon working would be really handy for me when doing automated testing of software in VMs. So far though, the stuff that Microsoft has been working on looks like it needs a lot of work to meet the quality standards of what I would expect to be required to be accepted into the mainline branch of a major open source project.