A heap sort does about half its work while receiving input, and the other half while producing output. Under normal circumstances, it is expected to take more CPU time than a quick sort, but less than an insertion sort. Therefore, in queries where the input source is slow but not CPU-bound, half of the time taken to perform the sort is overlapped with the input. Similarly, in queries where the output processing is slow but not CPU-bound, the other half of the time taken to perform the sort is overlapped with the output. Also, if the sort processing terminates without consuming all of its input, then some of the work can be avoided entirely (about half in the limiting case where no output is consumed), saving both CPU and total time.

In some cases, such as when a SORT is quickly followed by a CHOOSEN, the compiler will be able to spot that only a part of the sort's output will be required and replace it with a more efficient implementation. This will not be true in the general case.