h2 is supported by all major browsers (http://caniuse.com/#feat=http2). It’s the CDNs/servers that need to catch up. That’s changing fast in the past year.
It’s unfortunate that other vendors are not taking a look at HTML Imports, but they’re a huge perf win (eg. https://twitter.com/ebidel/status/636702345674424320). Parallelization of parsing, resource de-dupe, HTML/CSS/JS bundling, look-ahead parser support, etc.
> Otherwise it would end up with worse initial load performance due to the orders of magnitude more requests
The # of requests wouldn’t change. The difference is have the brower’s native loading pipeline vs polyfilled (xhr’d).
> what is the chance of a user in this scenario having an up-to-date browser
Yep, I hear you :) Making things faster for new hardware doesn’t hurt either. This space is changing fast though! Part of me thinks we should be planning for the future…but I understand there are real-world needs today. I’m always a big fan of measuring a site’s actual user base before making technology decisions.
Alex Russell may have looked at Shop (shop.polymer-project.org) on older, non chrome browsers on a hindered connection for first load. After first load, you have a service worker and network drops from the equation.