For the reason stated above.  The proposal should impose no real burden on OS-A class implementations, while ensuring a reasonable upper bound on the lack of resolution/accuracy.  If anything the discussion from others on this thread is to implement finer tick periods.  And, as others have noted, other architectures also provide a relatively fine resolution "time" (meaning in the >10 MHz tick frequency).  Without trying to list umpteen individual use cases, I think it's fair to say that this is a generally useful - yet modest - requirement.

Greg