Why wasn't this functionality ever baked into the web? HEAD requests could return a SHA hash of the content, and browsers would check peers before issuing a GET for the remote resource.
additionally, content-addressing provides another layer of security beyond location addressing. even improperly cached information is as secure as your hashing algorithm.
Either that or game/code assets on large college or corporate campuses.
Yes, it was a pain whenever the origin server had misconfigured caching headers. The "force reload bypassing caches" shortcut (Shift-Control-R, or Shift-Click-Reload-Button) was always handy. The move to TLS everywhere, which killed these "transparent" (when they worked) caching proxies, made the web much more stable and predictable.
Some ISPs, on the other hand, did run "transparent" proxies in their networks, causing a range of issues.
HTTP is just inherently server-client, decreasing load on servers by sharing between clients was not a design goal.
Although IPFS seems to want to implement something like that, and probably much easier to do in a content-addressable way (like IPFS). So content has a hash, and as long as you trust the source of the hash, you can download the content of a hash from anywhere, even your drunk and slightly annoying neighbor that you don't actually trust, because you can verify that you got the right bytes after all.
Bittorrent does it for transfers already, and it works alright, so why not for the web too?
Sounds simple, but number of cases where this would happen is likely pretty low.