Since when is a 4-to-1 speed-up "nano"?

It is called micro- or nano-optimization because the operation being optimized is so small that it is likely to account for closer to 0.1% or less of the total run time (instead of the 10% you guessed) of most non-trivial scripts, not because Benchmark.pm measures a small percentage difference.

Even if it accounted for 10% of the run time (extremely unlikely for an operation as small as comparing two characters), a 4-to-1 speed-up means only a 7.5% total speed-up. In reality, you are more likely to get something like a 0.075% speed up.

I'd much prefer that it take 3 hours rather than 12

See, this is why this cult is so seductive. No matter how many times someone explains why this type of stuff is almost never very useful, people see "Four times faster!" and start drooling. Using your own (highly optimistic) numbers, a 12-hour run would take 11.1 hours (not 3). I mention this not because I think 0.9 hours is a short time (because the reality will probably be less than even a few minutes saved for a 12-hour run) but to show how "Four times faster!" takes over one's thoughts and one jumps to invalid conclusions.

Even if you had a script that did (well, tried to do) nothing but compare lots of leading characters of strings, you'd probably find that the overhead would end up adding up to a lot more than the comparison time. Benchmark.pm does a lot of work to estimate the overhead and subtract it.

I also think noone would try the slowest methods benchmarked here except in a rather desperate attempt to nano-optimize, so even the 4-to-1 speed-up isn't real. It is a near-to-4-to-1 slow-down of a desperate optimization attempt.