A third vote for MIDI. OS/2 is (was? I think it still IS) far superior to Windows when it comes to timers and time resolution is the key factor in MIDI handling.
We have/had a great MIDI sequencing software (MIDI Station Sequencer) and some interesting experiments in score editors (like my ABC and Mup ports). So it's a start.
I know someone who is working on this, but preliminary work only in his spare time, I did some work on this around 2000, the issues then were:
RTMIDI was/is hopeless, Timur Tabi had done a partial MIDI implementation and then stopped work on it, he promised to supply source code but that never happened so the 2 or 3 major projects that actually used RTMIDI had to stop development including MIDI Station and AudioStation, the code as it was did not even support MIDI sysex passing, it was so basic. It was also in retrospect probably the wrong way around in the way it was implemented, instead of implementing MIDI functions it should have passed them through and concentrated on controlling the basics of the protocol and act as a timing reference, much simpler.
It would make a lot more sense to start again with something that emulates the MIDI part of VST3 (not the VST3 specification but the control section only, it is a bit like MIDI++), and then have MIDI as a subsystem of that, no need to implement the specifics, let the application or plugin handle that.
The basic Multimedia audio subsystem is ancient, it has seen no work on it since OS/2 2.0, it needs to be looked at, unusually since it was one of the few things that had been written from scratch by IBM so Serenity Systems said they could actually get the source for that as well, but that never happened either. While it functions perfectly well as it is, or reasonably well depending on what axis you look at it, the issue is if you want to support something that did not exist at the time of its original like DSD (HQ 1bit audio), multichannel audio processing (for music making or home cinema) or very low latency audio (not needed for MM but for music making), you might as well start from scratch.
The other reason that you might want to rewrite the audio subsystem is that modern microprocessors have SMID and other DSP functions that did not exist at all when the original MM subsystem was developed, a simple module or plug-in architecture that would allow audio processing to take advantage of this without going through the complexities of what we have with current windows and Linux solutions that are all over-ambitious
All in all the audio portion is actually not that big a project, not compared to other parts of the system, but it needs someone that knowns what he is doing, or is willing to learn.