problems with ntpd sync on RHEL3 ES

[Gavin McDonald]

Hi all,

I administer a number of IBM eServers, all of which run 
RHEL 3 ES, with ntp version 4.1.2-4.EL3.1

The config files, (Sanitized & quoted At the end of this email,)
are identical on all machines, and identify two windows
domain controllers as stratum 2 time servers. 

When I start ntpd, by doing a 
 # service ntpd start 
on each server,  The machines all synchronize correctly, to 
within seconds of each other, and everything appears fine.

However, after several days, (more than 5, less than 10,)
the clocks on two of the machines are over five minutes 
out of sync from the other two.  If I restart ntpd, the
clocks immediately go back in sync.  This has been happening
through several "tweaks" of the config files.

Has anyone seen this behaviour before?  The fact that
restarting ntp brings the clocks into sync seems to
indicate that the configurations are working, so I am
truly stumped.  Any help would be appreciated. 

Regards,
Gavin McDonald.

Config files follow:

(00:52 hostname:~) grep -v ^# /etc/ntp.conf |grep -v ^$
restrict default nomodify
restrict 127.0.0.1
server 192.168.1.50 stratum 2
server 192.168.1.52 stratum 2
driftfile /var/lib/ntp/drift
broadcastdelay  0.008
keys            /etc/ntp/keys

(00:52 hostname:~) grep -v ^# /etc/ntp/ntpservers |grep -v ^$
win2kserver0.domain.local
win2kserver2.domain.local
clock.redhat.com
clock2.redhat.com

(00:53 hostname:~) grep -v ^# /etc/ntp/step-tickers |grep -v ^$
win2kserver0.domain.local
win2kserver2.domain.local