Well that is because these people who write standard books know more about the real world problems and assume that they are known to anyone.
In a computer system - at least from 2000 and till now, memory speed is the real bottleneck. That is even if we increase the CPU speed very much, most real world programs won't improve in runtime as the memory speed is the real bottleneck - imagine a neck in a bottle. So, for performance reasons we can consider "no. of memory accesses" as a yardstick.
Now, even with above assumption, in real systems we often have cache which works on the locality of reference.