Curious: how many iterations did you run of each benchmark and what was the variance?