Can we say that algorithmic speed-ups are sufficiently rare that they are usually impressive?