When would it be appropriate to compare the performance of two programs$--$ A and B$--$ by measuring their execution time on a computer, as opposed to performing algorithm analysis on their underlying algorithms?