I personally believe that HTML is fantastic for what it was designed for: Structured Documents. But the world has gone more interactive and I believe the ad-hoc solution of Javascript, HTML and CSS is not "rich" enough to provide for the users demand.
I like the language independent VM idea as long as it doesn't turn out like bloated Java bytecode.
If the goal was to create a new language well Dart has done that.
If the goal was to give web dev some of tools of app dev (static typing, classical restricted OOP, better speed) I don't think Dart was the right answer specifically because I don't think it's got much chance of getting adopted by other browsers.
The practical solution IMO would be the CLR. Why? Because if Google had chosen the CLR they'd instantly have brought Microsoft and IE along with them at which point Mozilla would probably be pulled in if kicking and screaming.
That seems to me the most pragmatic solution but sadly Google is unwilling to do that for whatever reason.
Why are we bothering with developing new languages when what we should be doing is developing a web standard for a universal bytecode + VM? Hell, use the JVM or the CLR, just as long as it doesn't tie me to a specific language. Dart may be better than JavaScript (I would hope so) but that doesn't mean it's a silver bullet. If we create a standardized bytecode, it doesn't matter what language produced it.
I've been thinking about something like this for a long time. However, the question came up in an interview or presentation with either Brendan Eich or Douglas Crockford (I can't remember exactly where I read this) about why Javascript won out over a VM solution (supposedly this is what the JVM was made for). The most salient argument against a universal VM was that maintaining backwards compatibility between byte-code versions and the competing eco-system of languages would have been a nightmare. Instead the web targeted a single language (and all the trouble that has been along with it). I would rather like to see other languages on the web but I don't think a browser is good place to do it.
Instead what I'd rather see is http://www.cs.cornell.edu/home/ulfar/cuba/paper/ -- work could be done to improve that initial design, but the idea IMO is that the OS already handles processes and allocating memory well and the browser vendors are terrible at this task by comparison.
Either way, Dart is an interesting experiment and I'd give it a shot on some prototyping projects.