Hi! I'm the author of that benchmark. Note that the machine has been set-up so this particular comparison could happen without interference of other features which speed up performance dramatically, most notably the query cache. Also, regardless of the overall performance, what I've been trying to observe are the results when compared to each other.
That said, if you would like, I can post the scripts I used for the benchmark, I'd love to be pointed wrong as based on these benchmarks I've neglected this tmpfs idea.
Cheers