Native applications is a relatively fragmented market of different hardware and OS for platform, made more complicated by relative lack of interest (which is because the market is fragmented, a catch-22), and factors like needing to learn another programming language when you already know JavaScript and how it works on the Web, which is taught to more people every year for obvious reasons. Which is all why Github Electron, essentially a Google Chrome married to Node.js, both _JavaScript_ platforms, made such an impact when it was released. There's zero-install on the Web, too -- just follow a link and you're surfing applications. Python+Qt applications have to be installed, even if that means downloading these -- there's plenty of hosts configured to deny the user the privileges of running software they downloaded, no matter how native and how well mannered it is otherwise. There's fewer pairs of hands on the job (part of catch-22), and there's more standards and APIs to deal with, due to the fragmentation, even for all the cross-platform offerings. All this no doubt contributes to the market staying behind the juggernaut that the Web has become.
Before you roll your eyes and label me a millennial who's not seen anything but the absolutely appalling Web applications of yesteryear, fresh off inexperienced hands of developers who think they invented caching and what not -- I started off with x86 assembler and C then C++ in early 90's, and I hold genuine interest in everything we learned since before Intel made 8088 -- but I am simply describing the reality I see, not necessarily reality I want.
You're drawing a border on water -- there's no need to "separate" the Web from native. The Web is an application platform developed from a hypertext network (the old Web I re-label for comparison's sake), and the platform has tremendous value. You need to have tunnel vision to want to put genie back into the bottle, but again -- I absolutely hear and understand your argument. Do you have realistic suggestions?
Drew DeVault suggested another protocol, Gemini, a while back, having become frustrated with much the same observation you did. Just text mark-up served with efficient text-based protocol -- essentially a regression back to HTTP and HTML anno 1995 (possibly with more semantic elements). I think it's not only a fantasy but also a poor idea -- not because it's a bad idea in itself but because it assumes there's no possibility to do any of it with today's Web, but there is -- it's just that everyone's reaching for the fancy and the flashy once they start coding. What you were referring to with "focus jump around" and "shifting layout". We're sacks of flesh driven by hormones -- that's the best reason I can give you why the same platform that allows you to slap [a HTML that's worth reading](http://motherfuckingwebsite.com/), possibly [with a simple stylesheet that does the bare minimum to improve user's experience](http://bettermotherfuckingwebsite.com) -- is _not enough_ for authors. I'd call it "author's prerogative" -- the person who pays for the domain and the hosting, wants to exercise their authoring power and gets carried away with all the bells and whistles they slap on their pages. Users pull their hair out, in silence (or mostly ignored because "do I paint the walls in your house?").
Anyway, this is getting long -- the gist of my argument is that technically the Web is capable of supporting all the static HTML without an ounce of "shitty" scripting that makes everything border on "unconsumable". You're making a "dictatorship" argument along "if you can't make good readable sites, we're going to neuter the platform". But the platform _is_ what drives adoption of the Web, I say, albeit now nearing some cancerous growth from a skeptic's perspective. And yet: fix the _content_, not the _platform_. "Native" is just a word -- there's no native, everything is translated or compiled one way or another, including JavaScript (which _I_ consider a relatively bad general purpose programming language, even under ECMA oversight which fixed a lot of its warts, admittedly). Unless you're one of those ["real programmers"](https://xkcd.com/378/).