Good news here about the timing issues. Jan Wassenberg has finished the HPET driver and I am currently testing it here. It is working fine, great in fact. Just about to try it with actual audio generation.

I started by testing how it measures time compared with the other timers.

It is giving interesting results, here is a run I just did comparing all the timers measuring the time with them all roughly every 50 ms.

The high precision timer is the current default for Bounce Metronome and is the best performing one apart from the HPET. But it also is still glitchy with multiple millisecond glitches, though not as bad as e.g. the multimedia timer.

TECHY DETAILS

Here is a table showing several of the timers on Windows together with the new HPET one, on my laptop with turbo boost. MM is the multimedia timer which most music software uses. HP is the high precision timer. 

There will be a spread probably of a fraction of a millisecond between the times because of course youcan't query all the timers instantaneously, just as quickly as possible. The new HPET driver however is able to access the HPET time and return it to the calling application in around 3 microseconds typically so is very fast.

In this run then, assuming the HPET timer is the one that gives the accurate time, as seems pretty likely, then most of the other timers say that more time elapsed than actually did, overestimating by over 40 ms. 

The jump happens in this line here for all except the HP