Does it use SIMD instructions? Some newer cpus can do up to 16 comparisons at once. That also might mean skipping will be branchy and not exactly divisible on word boundaries for best vectorization.