There is a nice (but tricky) way to implement a high resolution sleep function on Win32. I checked the current APR code, but could not find such an extension to the standard timing functions.
Is it planned to implement that feature? If so, I might be able to contribute an unofficial raw version of that implementation - which in the end is a combination of multimedia-timers and critical sections that bypasses some of the scheduling problems.
Btw: How accurate is nanosleep on current Linux kernels? Do you get up to 1ms accurate sleeping times?
PS. ...just found APR and now I'm evaluating it as a replacement for our existing Win32/Linux API wrapper - so please be gentle upon newbie questions.