Is there a way to have constraints evaluated in a parallel/vectorized fashion like it can be done for the cost function by setting parallel_objective arg of fmin() to a callable?
There are plenty of problems where all or most of the computation is spent on finding a feasible solution.