It is a similar situation to legacy Fortran code in the academic physics community. Fortran is used because some guy wrote the core algorithm 30 years ago in Fortran 77 and never wrote any docs about it. "If it ain't broke don't fix it."
I don't agree with this mentality at all. In fact, I make use of Python almost exclusively for the work I do (in a small fund). My NumPy/SciPy is not good enough though to choose it over C++ when I need to perform some hefty number crunching.
How did you find using SAS/SPS? Is R not a good candidate for what you're doing?