The Python test is indeed a large outlier! I wonder how reproducible it is, and what the reason might be.