We should run benchmarks with bounds checking enabled to better reflect real world results. Yes, it might "lose" to C
Like for like comparisons are the best approach, making it clear what a given result is for. The most effective story for D is that it's fast.
