I think that if your entire program is written to heavily rely on polyfills in all areas then you may be asking for trouble. However, I think that extending objects through their prototypes is a valid way to implement functionality (as it's a defining principle of the language), and I don't think I would tell someone not extend the base String object if they invented some new great desired feature for Strings. So isn't that what a polyfill does? Regardless if the feature is implemented in 50% of browsers or 0% of them, there will be edge cases, we all know that. Polyfills apply a pressure to the industry that breathes movement into the language and makes people aware of the what is coming.
If you need a polyfil to accomplish your goals then I say use it. If there is a safer more reliable way to create the functionality then that's the way to go. The only bit about the ponyfill idea that I don't like is that when a feature does become supported, now all of a sudden I'd have a pile of code relying on heavy refactoring where polyfills can ease the transition.
Interesting article though, awesome, thanks!
This is the premise of Closure Library, e.g., goog.string.trim().