I think you are missing the point. The larger the scale is, the more things may be optimized and the more you need to be careful to focus on the important "3%". That said, it doesn't hurt to analyse in advance which parts might be critical and steer the architecture based on those observations.

Statistically speaking, 99% of game devs won't ever make a game like that. And if they do, they will probably know ahead of time they are making it.

100% of hobby game devs do it as a hobby. The learning process is the fun part. Which is also why 99% give up on their game when they see more learning potential in another area.