Currently we are working on a way to make 2to3 compliant with Django's doctests... but I have an alternative:
You know how in Martin's code you have a convert function that utilizes 2to3? After call 2to3 why not call a new python script that manually removes the unicode prefix from each string since Python3.1 only has unicode.
See Martin v. Lowis's comment in issue #4, doctest conversion should be done through setup.py using the 2to3 library
Running 2to3 with the -d flag would actually format the docstring as well and theoretically should solve some of the unicode problem, but since we should integrate with the setup.py then the recommended approach is to use distutils