If you take samples from different populations, then averaging those sample averages provides you with worse information, not better. The populations of “Oregon trying to score” and “Oregon with their backups in not trying to score” are not the same, so averaging sample averages from samples of each does no good.

In other words, Oregon’s YPP from the entire game does not accurately represent BYU’s defensive performance.

Your superficial analysis would indicate that BYU did better against Oregon than Wazzu did. But Oregon had to try to score the whole game vs. Wazzu. They didn’t against BYU. Thus, if you want a truer comparison of BYU and Wazzu, the part of the BYU game where Oregon was actually trying to score is a better basis for comparison than the whole BYU game.