This may be opening a can of worms, but I wonder if what we should really do is re-engineer regrtest from the ground up, keeping the existing regrtest around until we are satisfied with its replacement...and maybe some of the options wouldn't make the transition.  (I've used --single, but it's been a long time, and I think it may have only been when I was testing regrtest after modifying it...)

We could then shift buildbots over to the new command gradually (or, do a few first, and then all the rest).  Then when we're satisfied we could change -m test to use the new interface, but keep regrtest around, unmaintained, for a couple of releases for those who are still using it.

I haven't looked at Victor's code to see if I like his re-engineering, but I'm really talking about starting the re-engineering from the API, and only then thinking about the code to implement it.