"Utter elitism" is a nice article about this phenomenon

For that earlier section, we used smaller models trained on $S_4$ intersect $A_4\times 2$ (4,000 parameters) instead of $S_5$ intersect $A_5\times 2$ (80,000 parameters) -- the only reason for this was to allow for a larger sample size of 10,000 models with our compute budget. All subsequent sections use the $S_5$ models.