If you've got some scientific papers handy, I'd love to see them!
The point I'm trying to make is that YouTube has an incentive to design their system to not let traffic travel further than it has to (users closer to a data center hosting the content they want will get it faster). They build data centers close to where their users live. Even then, delivery is likely less energy-intensive than video transcoding, meaning large, specialized data centers make a lot of sense for that task. They then distribute transcoded content to smaller, regional servers to improve user experience ... again, specialized systems for a specialized task.
This means that YouTube has already distributed their system across many different servers in many different regions around the world, so in many ways, they already take advantage of the efficiency benefits of p2p, but they can carefully coordinate to reduce overall costs in a way that p2p can't (yet).
But the Fediverse will lag in efficiency for exactly the reason you pointed out: it's running on low tech, general-purpose hardware. Energy usage has the largest environmental impact by far. Hardware that is specialized (like Application-Specific Integrated Circuits) or newer will always outperform general or old hardware.