et lux in tenebris lucet

An official fix has been merged into Postfix for spools that are mounted with the “noatime” option:

– On backup MX servers where the queue file system is mounted with “atime” (file read/execute access time) updates disabled, the flush daemon would trigger mail delivery attempts once every 1000 seconds, thus rendering the maximal_backoff_time setting useless for backup MX service.

This is likely a result of my “Backoff not working?” thread on the postfix-users mailing list. (Although I’m still going to recommend leaving atime enabled, as I mentioned earlier.)