Sounds like the pmax port is pretty clean, interrupt-latency wise.
Here are some additional data points:

 -----  chronos.clock.org  ----- Sun 3/60 (NetBSD-current)
offset:               -0.004284 s
frequency:            12.014 ppm
poll adjust:          30
watchdog timer:       689 s
 -----  solar.clock.org  ----- SPARCstation 2 (SunOS 4.1.4)
offset:               0.001198 s
frequency:            7.310 ppm
poll adjust:          0
watchdog timer:       13 s
 -----  digital.clock.org  ----- SPARCstation LX (NetBSD-current)
offset:               -0.000544 s
frequency:            53.935 ppm
poll adjust:          12
watchdog timer:       12 s
 -----  atomic.clock.org  ----- SPARCstation Classic (NetBSD-current)
offset:               -0.000759 s
frequency:            73.927 ppm
poll adjust:          -18
watchdog timer:       42 s
 -----  cesium.clock.org  ----- SPARCstation LX (NetBSD-current)
offset:               -0.000472 s
frequency:            83.486 ppm
poll adjust:          24
watchdog timer:       19 s

Offhand, I'd say that NetBSD/SPARC on sun4m needs some help. That, or the
clock hardware in those sun4m systems is just ten times worse than its
predecessors in the Sun 3 and sun4c. I suppose that I ought to run
NetBSD-current on the SS2 for a week or so just to see if the problem is sun4m
specific, or endemic to the SPARC.

No Macintoshes running NetBSD here yet, but I have high hopes for
NetBSD/PowerPC one of these days...

	Erik E. Fair    fair@clock.org