The readme says "~50%" of the time, which seems to be a number pulled out of a hat (to be nice) rather than any serious attempt at quantifying performance (ther

The readme says "~50%" of the time, which seems to be a number pulled out of a hat (to be nice) rather than any serious attempt at quantifying performance (there are no metrics of any kind and the authors are asking for relevant benchmarks). It's more like the authors' feeling, than anything they have ascertained in any reliable way. That's pretty much the same way homeopaths test their "remedies" (i.e. water).