Resample only if too few viable points left

Even when the number of reduced parameters is over given thresholds, don't re-sample if there is still enough viable points left. Should significantly improve running-time for highly reduceable models like the one in the kuepfer-tor example.