I also disagree that the app-ification was an historical accident; I think it's the direct result of economical forces, and I'm not convinced that an alternative system could become popular while avoiding it.
You might want to look through the inspirations part of the doc to see what's prompted this. I've also turned up a number of similar projects after having started writing that essay.
But:
* The /proc and /sys filesystems on Linux
* The /net virtual filesystem on many Solaris systems, generally tying to an NFS network via automount.
* Various Plan 9 elements, including /webfs
* The Midnight Commander virtual filesystems, including remote access over ssh.
* SSHFS.
* Tools for on-demand and responsive media response. Tracking an element down to its source, and requesting, say, PDF or ePub rather than HTML might automatically provide those. The idea of being able to run, say, 'ls' on an HTML target and have it return a list of links rather than its full content.
* Unix mail, mailx, mutt, and especially mh, which provides a shell-based perspective to email management.
* Numerous document markup and preparation systems.
* Squid, dansguardian, Tor, and other Web proxy tools.
* Libraries and catalogs, generally.
* The Internet Archive.
* HTML doc "reader-mode" tools: Readability, Pocket, Instapaper, Outline.
* Search
Again: wiring much of this into the FS layer, abstracting away the browser, treating doc requests as queues, splitting up docs, media, apps, and commerce, integrating into the user's workflow rather than the publishers.
If it's not for you, that's fine. But that's where I'm coming from.
A filesystem is a CRUD (or BREAD) tool, based on standard metadata and interfaces. Part of that is making available extra and user-useful capabilities.
Another part of the problem is that the current concept of a browser, which lacks the idea of downloading-and-storing, on an accessible and persistent basis, online content, and emphasises application and interactive functions over document and reference capabilities, is ... something of a historical accident.
Why make use of some remote pinboarding or bookmarking feature if you've got a common URL referencing system, and the ability to capture, transform, rate/rank, search, and organise into a workflow, project, or support context the information you're accessing via the Web.
And to do that by way of tools which treat this as information-on-disk, either known and extant utilities, or extensions / wrappers / enhancements of these. Most specifically, by unwrapping online content from the browser context specifically.
This goes deep into principles of application, information, and interface design and use. I'm calling the extant model a dead duck.