Hello Mike,
The logs I sent were for only 4 operations (you can grep by "Operations uploaded") so 28 minutes is a very long time IMHO.
Moreover the behaviour is not always consistent, as I mentioned earlier if I cancel the job and start a new one with the very same operations it takes only 10 seconds (90% of the times, but then I cancel and retry again). It's as if I was being served by a slow server from time to time.
I think the auto-retrying and rate-limit logic is great, but I think this issue should be investigated before v201509 is sunset.
If there's no such thing as a limping server maybe the scheduling logic could be reviewed, IMHO a job with few operations should be executed much faster than a job with thousands of operations but from what I see this is not the case oftentimes.
I am surprised that only few other people reported this issue, perhaps many are still using the older version.
I also observed that there are days where the jobs are very slow (like today) and others where the jobs are only somewhat slow.