Yeah, this is sort of the problem with a percentage-based approach. You are better with a "pick the top 3 implementations" in most cases - that bags you Webkit, Blink, and Gecko in the browser example, and since we're ignoring the long-tail in either case, that's probably good enough