Don't worry, I had that happen on mine as well, but in every case, the my was a percentage faster and the ref was a percentage slower and everything else just hummed around in the middle. The point wasn't the benchmark. The point was that declaring aliases or setting up your own constant subs, are not that much slower than my'ing a variable or local'ing a variable.
The point of benchmark is ballpark, and all of the types are in the ball park.