Moreover, with the micro-optimized SIMD quicksort algos that are perennially cropping up on this website... I would be willing to bet that "sort and take first N" is objectively faster than my crappy Python implementation -- even if it is linear time.