Delayed job processed more than once

It's a bit hard to reproduce, and I'm not 100% sure what to share exactly.

Setup: We have one large redis in the middle, and a variety of different queues and workers connected. Each queue is being fed by at least 10 workers, and processed by another 10.

Now, I do have one specific type of job, that's enqueued with a delay (e.g. 70 hours). This job is also a bit longer, takes on average 1.8 seconds. And this is regularly being processed twice in parallel, by different workers, maybe even on different machines.

For example, I log the start of the job and got:
```
2017-11-16 10:56:12.733 - Job 2494621079...
2017-11-16 10:56:12.180 - Job 2494621079...
```

Is there anything I can do about this, or are we just cratching the top-line and may need to move to another queue system?

No source files, tests, or entry points are named. Start by reproducing the delayed 70-hour job with the described shared Redis, multiple queues, and parallel workers, then trace delayed-job promotion and worker processing; done means identifying the duplicate-processing cause and verifying that the same job is not run concurrently.

Written by the indexing model from the issue text.