I guess I'm thinking about this from the perspective of someone trying to make a computer for the general public. How will you convince anyone to buy and use a computer that can't browse the web?
(And I doubt this is a problem that will go away with time; the web is big enough that it seems unlikely to go anywhere anytime soon.)
To provide a little hope, I think it should be pointed out that the Frank project (VPRI) achieved web browsing that is compatible with the existing web protocol, using real objects. What it allowed is an extension of browsing on the web. So, it's not as if the two (good architecture and bad architecture) can't both exist in the same space, and be used at the same time. I think the key to answering your question is asking will people in the general public find the impetus to understand the limited nature of bad architecture, and (on one possible track) either use the good architecture to make the experience better, or (on another possible track) come up with a better computing/content architecture that does away with the web as we know it altogether?
Several companies making game consoles have done it repeatedly. Phone companies do it. People were actually buying dedicated, word processors for a while despite existence of MS Word & Internet. There's devices that only let you read books. There was one, popular computer that could only play about 5GB of music.
Seeing the patterns connecting them all? That it will be a niche market to begin with doesn't mean there's no market or it's not worthwhile.
Here's one for you given success of gaming and entertainment products: a all-in-one computer combining rapid iteration, memory safety, efficiency, and HW acceleration for common things (esp graphics); a Python or BASIC (eg DarkBASIC) designed for gaming w/ libraries for common features; a port of a game creator program plus examples & artwork to draw on; tutorials a la Realm of Racket or Land of LISP that teach you the language with successive building of game modules with increasing complexity or knowledge required; ability to live patch & debug a la LISP the games with failure isolation so no lost work or long times between runs.
Think people would buy it? Especially people new to programming who would find C++, Java, web stacks, and so on daunting with low-reward steps in the learning process? Could such a HW/SW combination be a 180 for them in motivation and learning?
(And I doubt this is a problem that will go away with time; the web is big enough that it seems unlikely to go anywhere anytime soon.)