I recently took the plunge and switched to using Python 3 on a daily basis. I found it to be surprisingly straightforward. As a newly minted Python 3 user, I am now licensed (perhaps even obligated) to nag Python 2 users about switching. Honestly, for scientific computing there are not yet any killer features in Python 3 that make a switch tempting. However, there are still some good reasons for scientific practitioners to switch. Here are five of them:
1. You will have to eventually.
Python will be around for a long time. Python 2 will not. You'll have to switch eventually, so why not do it on your own time, well before Python 2 reaches end-of-life support in 2020?
2. It's not that hard.
Most numeric-heavy code (e.g., without much string handling or text I/O) will require few and straightforward changes. If you have comprehensive tests for your code, switching to Python 3 is really just a matter of running the tests and looking up what needs to be changed for tests to pass. If you don't have tests, you probably developed the code by running it and seeing if it worked. In that case, run it with Python 3 and fix things until it works. If you are switching permanently, you can use the 2to3 tool to make most changes automatically. If you need to support both Python 2 and 3 simultaneously, the six library has made it relatively painless.
3. At this point it won't get much easier.
The majority of widely-used third party packages support Python 3. In particular, all the major packages for scientific work (NumPy, SciPy, Matplotlib, etc) have fully supported Python 3 for some time now. While it depends on what you use, the likelihood is that all packages you use already support Python 3. If you find one that doesn't, open an issue (or add a comment on an existing one) in the issue tracker for the package.
4. Python 3 will soon become better supported than Python 2.
This one is a bit of a prediction. As more and more developers switch to using Python 3 on a daily basis, new features in third-party packages will become better "tested" (through use) on Python 3. This is particularly true of smaller packages without full test coverage.
5. Be part of the solution.
The Python 2 to 3 transition sort of sucks. The sooner the vast majority of the community transitions, the sooner we can all stop spending time stuck awkwardly between two similar but incompatible languages. Developers can spend less time dealing with the dual support and more time developing awesome new features.