There is "acquisitionport 123" in the config files. I think that
prevents the implemented workaround from working. It can monitor only
client-specific sockets.
Try commenting that out and see if it helps.
Aha! That does indeed get us to 99.9+% hardware timestamps. Thanks!
We have that setting in place for semi-obscure firewall-related reasons but we may be able to get rid of it. If we do need to control the source port, am I correct in assuming that setting the "port" option on each "server" directive would let us do that without triggering this problem?