Observed and Documented Eventual Consistency in Amazon S3, OpenStack Swift, etc.

A more interesting and accurate test would be to perfectly synchronize two clocks in different portions of the world, and then run writes and reads on predefined sets of keys from those servers and log all times carefully. You could then merge the logfile to find inconsistencies (couldn't read X at time 100 but it was created at time 95).

If you're testing from only one computer/connection, you're really just testing which service most reliably routes you to the same backend servers. Consistency is a lot easier if you already know everyone who will make the request and can aggressively cache on the single specific frontend server.

I suggest you to be more clear about what the results you are showing mean, so one doesn't have to dive into the source code to find out.

Good work.