There is no meaningful performance increase to go to a backwards incompatible version? Three letters: DOA. if not performance then at least we'd need some crazy new feature like good multi threading or perhaps running on a new relevant platform (say ios or android). Otherwise we will be 2.X forever.
It's only when you start doing unicode seriously that you see how broken python2 is. It is not unworkable - but it's a mess that is hard to resolve, and keeps biting you in old code paths that weren't tested with character from this subset rather than that subset.
So, Python 3's killer feature is reasonable Unicode support.
The problem is that reasonable unicode support is a non-issue in many areas were python had a strong competitive advantage over other languages (e.g. data analysis, scientific python).
Actually, there are some minor performance increases, but there are tons of minor improvements, and the new async framework landing in 3.4 is very exciting.
