And of course simply linking two documents together isn't that useful, you have to say WHY they are linked. I.e., the semantic triple (https://en.wikipedia.org/wiki/Semantic_triple) of subject–predicate–object, or maybe more informally you are simply saying X relates to Y because of Z, where Z is akin to the predicate.
Currently in HTML hypertext we're stuffing Z into the link text, which sometimes works nicely and sometimes works very poorly. But in an external document you have all the space you want to explain the relation between the documents.
Obviously there's lots of shortcomings of adding a new document to the web to explain every relation between existing documents. But I think it's a good starting point. We're missing things like:
1. Reliable deep linking to documents. We have ids, YouTube timestamps, etc., but finding these is an ad hoc process and they aren't always available.
2. Widespread transclusion tools. We actually have some now, in the form of link previews or OEmbed. When you post a link in a comment or post on Twitter or Facebook, they effectively transclude the link into the document. Not fully interactive, but it might be a better balance between linking and viewing than traditional/literal transclusion.
3. Discovery of these annotations or commentary. There's a hard CS problem here, to maintain privacy while also trying to find serendipitous results. Maybe it involves pre-loading lists of documents from the locations you want to "discover" from. Maybe it requires some understanding of privacy levels, or whether content is personalized or public. Or we use the technique we have now: lead with commentary, with no attempt to discover it after the fact. I.e., I know there are comments on https://www.reinterpretcast.com/open-hypermedia at https://news.ycombinator.com/item?id=17690865 because I found the document on https://news.ycombinator.com/news – is serendipity even a thing in a place as large as the web?
4. Maybe publishing tools... do I want to post a Tweet to describe every relation I see? But maybe I do, because even if organic discovery is possible I probably also want to publish a feed of my own annotations, and I want to be part of a community of people doing this, and Twitter is a reasonable example of this.
5. Some sort of representation of these links when they've been found. Even without fancy discovery this is necessary. Right now if I click on a link from a post like: "OMG this is the stupidest argument ever: http://example.com/some-stupid-document" it will look like any other page I've opened. Only if I remember well why I clicked on the link will I understand that I've been offered something with derision. The browser has to do something here, all it has currently is the back button to understand why you've gotten somewhere (and that doesn't even work consistently in these cases).