My favorite tool for measuring this sort of thing: NTP

 -----  solar  -----
offset:               -0.002539 s
frequency:            7.636 ppm
poll adjust:          -12
watchdog timer:       19 s
 -----  digital  -----
offset:               0.001037 s
frequency:            54.117 ppm
poll adjust:          30
watchdog timer:       22 s
 -----  cesium  -----
offset:               0.000424 s
frequency:            82.791 ppm
poll adjust:          0
watchdog timer:       60 s
 -----  atomic  -----
offset:               0.002373 s
frequency:            71.968 ppm
poll adjust:          -30
watchdog timer:       45 s
 -----  chronos  -----
offset:               -0.001247 s
frequency:            12.063 ppm
poll adjust:          30
watchdog timer:       472 s

solar - SPARCstation 2, SunOS 4.1.4, essentially idle
digital - SPARC LX, NetBSD-current, June 18, NFS server
cesium - SPARC LX, NetBSD-current, August 20, user system
atomic - SPARC Classic, NetBSD-current of yesterday, NetBSD build system
chronos - Sun 3/60, NetBSD-current of yesterday, NetBSD build system

The thing to look at is the frequency - that's a measure of how badly the
system is drifting, smaller is better. Such drift is a measure of two
things:

1. the quality of the clock crystal used in the system

2. the OS interrupt latency for clock interrupts

I'd say, off-hand, that the numbers above mean either that Sun started
using cheaper crystals in the to sun4m than they had in the sun3 or sun4c,
or that NetBSD/sparc has an interrupt latency problem...

	Erik <fair@clock.org>