Comment by whizzter

Think of it this way, the global timer resolution of the system is minOf(allProcessesTimerResolution). If no process needs higher accuracy timing then there is nothing hindering the system from sleeping longer periods to save power and/or have less interrupt overhead (An feature I'd say).

These API's are from the 90s, in the beginning of the 90s where these API's are from having an global system interrupt firing 1000 times per second could very well have taken a percent or two or more from overall CPU performance (people already complained about the "overhead" of having a "real OS").

On the other hand writing audio-players on DOS you had the luxury of receiving your own interrupt within a few samples worth of audio, this meant that you could have very tight audio-buffers with less latency and quicker response to user triggers.

Not having that possibility to get that timing fidelity would have made Windows a no-go platform for audio-software, thus giving developers the freedom to enable it when needed was needed. Removing it in the next 10 years would probably have risked bad regressions.

Like a sibling comment noted, they finally removed it during Windows 10's lifespan and with modern CPU's _AND_ multicore they probably felt safe enough with performance margins to separate high accuracy threads/processes to separate cores and let other cores sleep more and actually win more battery life out of it.

It might not be "perfect engineering", but considering the number of schedulers written for Linux over the years to address desktop(audio) vs server loads it was a fairly practical and usable design.