When I run Berkley delta, I often see that success results are clustered, i.e. there are sequences of success results interleaved with sequences of failures. I think your strategy might be suboptimal for this case.
Yes, but since failed tests significantly outnumber successful ones, I'm guessing the current strategy is probably good enough.
Also you probably should switch to sequential execution when predicate becomes fast (e.g., faster than hard-coded or configurable threshold)
Definitely. John
