CAS after setting clock forward and then back is not correct

There is a test which:

• 4 node cluster, time stamp bucket
• sets the system clock on one node ahead one hour
• does 10,000 sets
• gets a CAS, it reflects the clock being one hour ahead (poisoned) - 0x148EAA59CC0E0000
• move the clock on the one node back one hour
• do set for a new key and get the CAS - it is much less than the previous CAS 0x10016CA5AAA8E14

There is a repro environment:

   ssh root@172.23.105.57 

   cd /root/eric/testrunner

python testrunner.py -i hostname.ini -t epengine.lww_stats.LWWStatsTests.test_poisoned_cas,lww=true

If I run the same test from my Mac to the same VMs the problem does not occur, a difference being my Mac has a larger network delay to the servers.

Logs are attached, let me know if there is more information I can provide.