> Presumably DELAY UNTIL is
> implemented with an
> interrupt. Could there be an overhead of some 10 or more
> microseconds
  I would be astonished if Windows did short delays with interrupts.
From my timings, it takes 4.9 mics just to do the OS call etc for "T :=
Ada.Calendar.Clock" (and since Gnat apparently uses the same (the only)
clock for Real_Time and Calendar, that would be same). Most Windows
calls specify times in milliseconds (though that is often misleading
about actual precision).  The thing that ticks at .8 mics is accessed
via a "QueryPerformanceCounter" call - I don't think they really think
of it as a general purpose clock at all.
> So would be surprising in a
> realtime
> operating system, though it is not surprising in a commercial
> operating
> system.
  My timings are running on Windows 95, which is the latter.
Does any version of Windows actually do timing and interrupts in 1
microsecond?