You can run <script type="text/coffeescript"> today. All of the interactive bits on CoffeeScript.org are implemented with it -- and here's an example of a site that uses it for something a little more creative (View Source):
This might not be a bad idea for a future project like a hypothetic HTML6, but I still think it's a bad idea _in general_. The web has consistently been about standards, and enabling developers to write code once that runs on many platforms (although the realities of browsers' quirks means that you will need to endlessly tweak your code to work on a given platform). Were we to allow developers to write in any suitable language, there would be much much more work on the part of end users and browser developers to make sure that their system is up to date with the latest language X plugin so that websites that use language X as a scripting language function well/correctly.
In short, this is a bad idea because it would fracture the web, when really we should be embracing standards that make the web simpler and easier for everyone to use.
Anyone sufficiently motivated and competent could write a browser plugin that holds a run-time for Python, Ruby, C#, whatever. The problem is getting people to install it or to get the browser makers to include it as a default. By historical fluke JavaScript got baked in to browsers early on.
Flash blockers are extremely popular these days.
We've had that already, and it didn't go too well:
<applet>
Also, until all current browsers (including IE9) die, any such script will be unusable or will have awful performance nad slow start caused by compatibility shim.We're stuck with JS, because it happened to be first one that's good enough. Cost/benefit of having multiple built-in run-times (that won't be more mature and interoperable anytime soon) or adding a VM is too large.
Maybe the next standard should require support for an additional language. The competition would benefit everybody. Even if we just put all the language names in a hat and pick one, we'd be on a better path than we are now.
The real problem is getting the language deployed on a critical mass of browsers.
You can do it today by compiling to JavaScript. See: Objective-J, CoffeeScript, Gordon (Flash), GWT (Java), Pyjamas (Python), ParenScript (Lisp), HotRuby (Ruby), haXe, etc.
But of course JavaScript isn't exactly the best language to target, and it could be a lot faster if there were a secure low level VM we could target. Something like the JVM, LLVM, or NativeClient. Then the language runtime could simply be downloaded the first time a page requiring the language is loaded.
(I don't think I'd do this myself, even though I like Python quite a lot.)
Also, most browsers don't run really run Javascript, they run ECMAscript or JScript. It's really jQuery converted to bytecode/assembly/C on SpiderMonkey/JagerMonkey/V8/Nitro. Sure you can create a high performance web server, but that's not really what V8 was meant for.
/pedantry
Not to mention RubySpec, which all major implementations are working to conform to. Even though that's built off of "What MRI/YARV do"...
It seems like any language that had appropriate DOM bindings should be able to run so long as the user's browser has the appropriate interpreter. (Kind of like a plugin, but for embedded interpreted languages.) For example:
<script language="Python" version="2.5+" downloadfrom="http://www.python.org/dompython ">
Is this just too much to ask?!