I haven't decided how I want to go about an actual response to that polars article yet, but here's the VERY unofficial numbers for the datasets in that blog post
SuperIntervals still wins in some cases, but returns results in reverse ordering. If that doesn't matter, SuperIntervals is the fastest