My primary machine runs Arch Linux, a distro that switched the /usr/bin/python link to python3 some months ago, thus generating a sort of "majority Python 3" environment. I myself, having been rather apathetic on the switchover before, have switched completely to Python 3 just because using Python 2.x has become a hassle.
Admittedly, I am in the fortunate position that all my active python projects could be moved over easily, once boost-python and matplotlib were solidly 3-ready. Other libaries are more problematic (looking at you, every Django-plugin out there).
My point is, Python 2 won't be good enough for long and for some of us it already isn't any more.
Citation needed. Even the devs aren't trying to sell it as much more than a clean up. Very little is actually qualitatively better, and there are some regressions also - for one, last time I looked it was still about 10% slower than Python 2 in most benchmarks.
(only a little kidding).
This isn't true at all. The negative is also often seen in the form of a null hypothesis. "There's someone in the fridge". "No, there isn't". Here is a case where we can prove a negative. Some negatives by their nature can't be proven, but as a blanket statement it's false.
You yourself mentioned that benchmarks were worse - this is proof of the negative, and you could have cited those benchmarks in your argument.
Also, even if I am starting an existing project, what if down the line I must use a library that's only available for Python 2? I am not going to rewrite something like that because it will cost my client money if I spend time on it.
The problem with Python 2 is that it's good enough. Python 3 is a huge improvement, yet until I can be reasonably sure that all libraries I am likely to use are available for both, Python 2 is the winner.
Having said that, any personal projects I have started recently have all used Python 3 and it's been great. I also try to make sure that any libraries I write work in both.