And while people can build crawlers that can fetch Javascript rendered pages, these scripts will have to drive real browsers with rendering engines, not just simple HTTP clients. And this is a problem, because fetching millions of web pages will require serious infrastructure work, versus a crawler using libev that can fetch dozens of pages in parallel without a sweat.
Also, that spec you gave is not exactly how Google operates. That's the advice Google gives for making web apps compliant, but the reality is that Google is smarter than that and it takes Google a good amount of effort to fix other people's screwups.
Accessible to crawlers doesn't mean Google-only and you should be careful not to throw the baby along with the bathwater.
Also, if search engine referrals is all you're interested in, if Google drops you from their index, your website might as well not exist, right? As I said, be careful not to throw the baby with the bathwater. Google is not what made the web great, but a Google dependency might break it and website authors can be blamed for this.
It's not a future-proof solution.
IE 9 comes close, and can be considered modern.
This is because the # in an URL is supposed to be an identifier of a element in the page, not an actual resource.
This sounds like a huge imposition to implement -- so much so that I seriously doubt anyone will bother for anything even moderately complex. Not only do you have to produce an API for the client-side javascript to call, but you also have to render static HTML for Google to index.
Your second point has merit. However I'd counter it thus: * You can load a HTML page first, containing a representation of the end content, so you can avoid the flash * You can show some sort of loading indicator on page load. Users are usually fine with waiting for the initial page load - it's only subsequent interactions that need be fast.
In other words, there's no reason why the page need be blank.