The reality is that, why would you ever render the same full-res image twice, once with DR and once without? This isn't something I can imagine anyone doing.
To answer your question, reality here is either:
"Cool, half of farm is idle, let´s do a quick DR so the client get´s the current state of the project today while I can check if everything´s ok when working on another one"
That is DR.
or
"Damn that dude on my right side is doing his bl**** animation sequence once again, farm will be busy for the next 10h. So let´s throw it in BB and get the result tomorrow noon."
That is single box rendering.
We do quite a lot projects iteratively with (expected) regular geometry/design changes. And a shot is almost always rendered fully even when only parts are changing. Then the result is just exchanged in the (non destructive) post - nothing unusual.
I´m now thinking of setting aside adaptivity until this is solved (even when I have to wait for v1.5) if I judge the situation correctly, because it will complicate things for us (different settings single vs. DR and/or noticeable different results).
Shifting from #passes to noise limit / sample error or whatever it will be finally called as stop condition may be a solution which I´m testing right now, assuming that the computed error level is the same with or without slave contribution. But there would be still region rendering where things may be completely different, let´s see.
(...)without getting our knickers in a twist.
Natives always force me to lookup something :)
Good Luck!