I can't help but think that people who want to replace HTML+CSS+JS haven't written a widely used web application before, because the idea seems so absurdly impractical and would become such an incredible time sink.
I'd tend to agree on those who want to replace HTML/CSS/JS with some other abstraction. What I'm suggesting is that they're arbitrarily divided and should be unified under one syntax, which then compiles into the served "bytecode." JSON would probably make the most sense. But, as you suggest, it would take some time to get it right. So I'm sticking to doing it by hand, all divided up, for now. Maybe after our startup launches...
I think even just MzScheme could become viable if the untyped stuff gets a bit more polished and there is just a lib everyone can agree on called "pragmatic.ss" that pulls in the relevant SRFIs.
Most of lisps problems these days stem from social issues and logistical issues. The core tech is excruciatingly sound, and the language itself is great.
To someone out in the startup circles, it almost seems like Lisp and Scheme focus more on being a standard than being a pragmatic and global platform. While noble, this is not the path to massive popularity. Some people might be okay with that, but it seems to me like you could serve both goals and end up with an overall better product.
CSS is very difficult to replace with an abstraction because it's too quirky. Compiling a high level language down to CSS without human intelligence is so difficult, it's completely impractical.
HTML, on the other hand, is easy to abstract. I'm glad I did it in Weblocks, it saves me an enormous amount of time. ASP.NET uses this strategy (use the DataGrid control for a while, and then try coding without it).
