I can see some interesting approaches here. Given n threads/workers you could divide the search space into n sample points (for simplicity let's divide it evenl

I can see some interesting approaches here. Given n threads/workers you could divide the search space into n sample points (for simplicity let's divide it evenly) and run the repeated test on each point. When a point hangs, that establishes a new upper limit, all higher search points are eliminated, the workers reassigned in the remaining search space. Given the uncertainty I can see how this might be more efficient, especially if the variance of the heisenbug is high.