Fix benchmark calculations

[mlrawlings]

This fixes #13 - the minified difference is being divided by the unminified/unoptimized time when it should be the minified/unoptimized time. This leads to incorrect results in the improvement column for all minified times. (looks like a copy/paste error)

I didn't actually re-run the benchmarks, I simply recalculated values from the table - because of this there is some rounding error, but the overall results should be more accurate than they were.

If you choose to just fix the benchmark and regenerate things, that's fine too.

[nolanlawson]

Thank you – this is why I do things out in the open: so that people can check my work and correct any errors. 😃

I'm not sure why the opt vs unopt improvements were slightly off as well, but I'll need to check my code and see if there was a rounding error or something. In any case, thank you very much for the fix – I checked manually and these improvement numbers are indeed correct. Cheers!