From the article it sounds like they're basing these tests off probably their dev machines. Seems like they should be trying this out on slower machines.
It sounds to me like they are testing it on slower machines (specifically slower hard drives), as they specifically call out one of the "smaller" optimizations as having a significant impact on machines with slower hard drives.
It sounded to me like they were extrapolating though, "This resulted in a ~100ms improvement on a fast machine with an SSD but, since most of this work was I/O bound, we expect it to be even more noticeable on slower hardware." This in fact was where I drew the inspiration for my comment from.