| For NTP, sub-microsecond resolution is very desirable. 

For all practical purposes (i.e., unless you are Mills or Lothberg),
making up a non-decreasing number for each call to a kernel time-stamping
routine made within the same microsecond or tenth of a microsecond,
would be more than adequate.  Mills and Lothberg probably lack a
practical purpose, but they do play around with hardware that can
actually detect drift on those timescales.  AFAICT, just about everyone
else who can, aren't playing with NTP.   (Except maybe the UTC(SP) folks,
but that's because their "algorithm" is an Earth Studies geek who
likes measuring continental drift on hourly timescales...)

How good do you think your quartz oscillators are at
various timescales?  

How to make a UTC time scale operator handwave: ask him or her
the following questions:

	1. why do their masers drift?
	2. who needs the resolutions they can get from their masers anyway?

Ah well... if it turns out the answer to 2 is "the NetBSD community",
and it leads to funding, everyone'll be thrilled.

	Sean.