I've worked in industry for 20 years. People take the same version of the interpreter, or start with existing scripts, to create new scripts. Even if there's a one rev difference, people often don't bother to upgrade. There will probably be hundreds of thousands, if not millions of Python 2.x scripts written over the next 6 years. Many will be small, of course, but it's code that will have to be checked against 3.x at some point. I threw out the conservative number because I thought it would be obvious to most people that given Python's popularity, a lot of new 2.x code will be written.

Obviously, if people were moving more quickly there wouldn't be a need to push out 2.7 support until 2020.