My guess is that this is a > which should be a >= or something. Mail is the only thing that has this additional supplemental check. It's no longer really necessary (long ago, mail wasn't part of the task queue, but we got that cleaned up and merged at some point) so a cleaner fix might be to just remove it entirely, although sending mail is a heavy enough side effect that it's not terrible to retain it as an extra check either, as long as it's not making a mess.

Hmm, is your MySQL clock about 20 seconds off of your PHP clock, maybe? You can check the former with:

SELECT UNIX_TIMESTAMP();

You can check the latter with:

$ php -r 'echo time()."\n";'

From a cursory read of the code, it looks like we're putting at least 15 seconds of buffer into the queue, so it's easier to imagine that this could be a clock skew issue than a race.

On reviewing the code, I'm leaning toward removing the redundant mechanism in MetaMTA, but verifying that it's a clock skew issue would give me far greater confidence about doing that.

No, my mysql and php (and everything else) are on the same machine, so there's no time difference there:

$ mysql -Be 'select unix_timestamp()'; php -r 'echo time()."\n";'
unix_timestamp()
1386268344
1386268344

...

Hrrm, troubling.

I'm just going to nuke the old code anyway, the new stuff is better on a lot of dimensions.