I don't know how these benchmarks work (do you do a hundred runs? A thousand runs?), but 0.1% seems like noise.