I'm sure we could have some sort of JSVM, we could even have some sort of bytecode for the HTML, but it would destroy part of what makes the web so great: it's openness. The ability to view the source of any webpage, and even make chances and enhance it locally.
There is also the fact that certain optimization can only be done by the browser if it has the actual source code rather than the compiled bytecode.
In both .net and java there exists excellent decompilation tools where you can easily decompile the bytecode back into java or C# or anyother language. It's not necessarily going to give you the original source code (especially missing comments) but it works incredibly well nonetheless. You could easily debug in it and edit it dynamically in browser tools as well. This is pretty much a solved problem. I wouldn't consider this a limitation worth talking about again.
Not really. HTML would remain HTML. We're talking about JavaScript, and that is already minified (and sometimes obfuscated) beyond recognisability for a human.
I'd be curious to hear what kind of optimisations become unavailable.
I think you're arguing a bit of a strawman when you mention viewing page sources: the discussion is about replacing JavaScript, not HTML. In any case, the fact that (non-web) applications are compiled into binaries hasn't stopped open source from flourishing.
There is also the fact that certain optimization can only be done by the browser if it has the actual source code rather than the compiled bytecode.