CSS has tons of problems, but almost every replacement for CSS that I've seen proposed would almost certainly have worse performance characteristics than CSS, optimally implemented, would. Most of them kill parallelism and kill opportunities for optimal GPU usage—they're proposed cures that are worse than the disease. The problem is mostly that browsers don't implement CSS as well as they could due to legacy, not that the technologies are fundamentally flawed (though this is not to say that they're not improvable, far from it).
To create a better architecture, you first need to have an in-depth understanding of what the problems in CSS actually are. Otherwise, you'll just trade one set of mistakes for another. (In particular, a CSS replacement that optimizes for Web developer productivity will not necessarily have acceptable performance characteristics—too often, people assume that "easy to develop in" implies "runs fast", which in fact these concerns are often in direct opposition.)
Surely there are some hiccups left in the design that come from its organic growth and could be smoothed out; but those are not as much as you would think, and most of that work has already been done in html5+ES6.
The end result is a platform that is great for zero-install distribution of web applications AND web documents. A new framework that centered only on the first goal would be severely limited, and a new one that served for both would look a lot like the current one.
Except the Web had already been invented at Xerox PARC and further explored at ETHZ using native technologies, so I guess some had already a pretty good idea.
The lackluster performance and high resource usage associated with web apps is a direct consequence of the complex patchwork nature of each component involved. Quite simply, the web was not designed to be used the way we’re using it, and evolving standards have made browser rendering engines horrifically complex (and in some cases convoluted).
What’s needed is a fundamental reimagining of HTML, CSS, and the DOM at minimum with little to no regard for legacy. In the 90s, we didn’t know what we needed but now the picture is crystal clear. Should we use this knowledge to design a new web, I’m positive that the result would be just as open as the web is now while performing tens or hundreds of times better — it’d likely be good enough to make web apps competitive with native apps under a far wider set of circumstances.