When a user stops watching for whatever reason, they leave the swarm. If a client has been getting segments from a peer that is close to them, and that user completes watching the video and closes the browser window, that peer goes offline. The player client then needs to find another peer to pull segments from, and hope that the peer can deliver at the same speed. Because most of these player extensions are on top of HLS or DASH, you're not "buffering" in advance very far, so you're sensitive to this bump (this is tunable, and P2P clients generally buffer more segments than those delivered by CDNs exclusively). Whether you can't find a peer with the segments you need, or the peer can't deliver fast enough, or the algo stalls and doesn't fail back to CDN based delivery fast enough, you get a playback stall.
Playback stalls always add frustration, and they're often multiplicative. One stall, you assume there's an internet issue, and it'll be fine. Two stalls, you grow frustrated with the service. Three stalls, you close the window. And now there's another peer dropping out the swarm, causing more interruptions in playback for other users.
IMO, as the tech stands today, P2P video delivery doesn't work super well unless you're setting up a P2P CDN, or have some other mechanism that allows for a peer to more gracefully exit a swarm to avoid stalls and signal to clients to shift before an abrupt exit.
Even then though, your average content viewer has different standards. We're opting for 1080p/4K content over the much more common smaller feeds. Netflix loads the first few segments of even the most esoteric content not in your local ISP cache in 100's of milliseconds, users flit from YouTube video to YouTube video randomly, and scroll TikTok like we used to scroll through the Twitter firehose back in the hey-day of the PopcornTime. More content is consumed on mobile devices than set-top boxes, which brings power and data constraints, both from metered data and with networks that are both highly latency variable and not tuned for upstream bandwidth, that make participation in a P2P swarm not ideal.
While content protection is certainly a core tenant of why content is centralized, if the big players at YouTube, Netflix, Hulu, et al, haven't figured out a way to shirk their highest cost center in lieu of putting it on the backs of someone else even cheaper than ISPs without causing a substandard consumer experience with P2P, then you can be darn sure that this is a tough nut to crack.
Also, even if you ignore the "we're camping next to fascists" vibe of the whole thing, PeerTube quality-of-experience management is deeply poor besides. If you're using some small DIY CDN, you have predictably bad performance, instead of suddenly bad performance, and if you're using a big CDN, you have working traffic steering.
If I had a website where I'd host my videos as an alternative to YouTube, I'd probably go with PeerTube and have people support me by seeding videos for me.
I tend to think that, since you're already foregoing discoverability by not federating with other PeerTube instances, you might as well also just build your bootleg CDN off of Hetzner nodes or whatever and have points of presence in places you care about. It's not that expensive and it's likely to be more consistent in perf. (You could say "well, run PeerTube and register your pops as peers", but other peers falling out of the swarm is the problem, not your own.)
I just don't really see a point to doing so as a small/individual creator or as a brand/channel.