Interesting. Looks like the reimplementation followed a different strategy. Makes me wonder if a C rewrite of the Python script would be even faster.