IMHO the problem is that you actually do need a decent chunk of all that stuff. Not all, but a good chunk.
A UI engine is a problem on par with creating a good game engine like Unity. If you don't have a loooooooong list of capabilities and polish your UI is a toy. By the time you extend and build it out enough to make it not a toy you will have re-implemented either Aqua/Cocoa, Microsoft's UI, or the web but yours will be new and nobody will know how to code for it.
Now make it flexible and programmable and extensible enough to be future-proof and battle hardened. Now make it work remotely over the wire with good performance. Now make it fast and efficient. Now make it secure.
The web isn't perfect by any stretch but it's IMHO the best combination of:
- Complete
- Modern
- Battle tested, including security
- Maintained by professionals
- Can be made to look good
- Has a yuuuuge developer community that isn't going anywhere
Qt is the closest other contender but its dev community is less than 1% of the web's. It's also starting to... ahem... look more and more like a web stack. It's still a little lighter and faster than the web but not much, and IMHO that gap is narrowing and will narrow even more. Look into Mozilla Quantum for an example of what's coming. We're going to have aggressively multithreaded GPU-assisted HTML+CSS renderers and web assembly soon.
JS is not my favorite language but it's not bad either for what it is (IMHO) and ES6+ features make it a lot better. It's a very acceptable "BASIC for the 21st century" and is perfectly fine for driving a presentation layer, schlepping data around, and controlling stuff. JS VMs are impressively fast given the difficulty of accelerating the language, and they are not overly heavy. WASM is going to break down this barrier as well by allowing CLANG to target the web as a native platform.
Web renderers are heavy but an OS that embraces the web as its presentation layer across the board could load one instance of all that bloat instead of one instance per app.
(A web browser on this imaginary platform would be a thin wrapper around an iframe.)
You'd have to pay a lot of attention to security. You might have to fork or implement a few added things into a good web engine to make it ready for such a use case, like better security models and resource management models around iframes. But that pales in comparison to the work it would take to re-invent a stack from scratch and reach this level of capability and then to build an entire ecosystem of this size and breadth around that stack. Forget about that unless you have a billion dollars to burn.