David,

Before spouting off, it'd be a really good idea if you took the time
to do a little research into NTP, and into Ethernet.
You will make less of a fool of yourself that way.

For Ethernet frames, I am talking about timestamping frames,
from multiple gig-e NICs or 10gig nics, *on a single box*.

A Tualatin-class i386 motherboard doing IP forwarding between multiple
gigabit NICs can quite easily keep up with streams where multiple
frames arrive within a single microsecond. A prestonia-class
motherboard has bandwidth to handle most of a 10 gig ethernet. If you
care about timestamping individual frames (and there are a number of
NetBSD users who do), microsecond-resolution time is, TODAY, not quite
good enough.

For NTP, sub-microsecond resolution is very desirable.  Poul-Hennig
Kamp claimed much better than that; Dave Mills argued that PHK was
measuring the wrong things and sent him a cesium standard (maybe
rubidium?) to check.