If you love cocoa so much you should checkout: http://www.cappuccino-project.org/
I remember when Swing was big. Everyone was gushing about it despite obvious flaws: non-native UI, poor startup time, horrible accessibility, and ridiculous memory consumption. Industry loves the idea of one platform-to-rule-them-all, no matter how far behind the uber-platform is. It would rather dump endless amounts of money and time into a patchwork of hacks than admit that it was wrong.
We're way past due for a VM for webapps. I don't see why the VM couldn't be added as a dual to the current HTML/JS/CSS model. URIs beginning with app:// would use the new VM model. There's a shocking amount of defeatism, both by developers and institutions surrounding the tooling of webapps.
I don't know what the answer is, and I agree with you that a huge part of the problem is that trying to build web applications on a platform intended for web documents doesn't work very well. But it seems to work better than everything else we've tried.
I'll be the first to say: creating views in those systems is usually not as easy as HTML. But the coherence of the tooling makes up for it quickly.
It appears the development ecosystem is optimizing itself around what lets you throw shit up as fast as possible, regardless of how coherent or painful it is to modify in the future. Time-to-market has always been a concern, but the notion of Internet-time has only accelerated things.
This also explains why R&D has taken a backseat many times; we often confuse it with new products (that use the same tech).
Then I started doing CocoaTouch for a project with a friend. I can't view the source, and the tools suck ass (xcode, obj-c) but compared to Javascript and especially the frameworks for web apps, it's wonderful!
The JS frameworks, all of them, make things a hundred times more complicated than they need to be. If web programming was like CocoaTouch, but with a dynamic language, it would be great. /rant