Python as a statistics workbench

Depends on what you're doing. In my research the moment I'm working on difficult to access data or with new techniques, Python becomes the better tool immediately.

R is absolutely beautiful for manipulating small data sets. It's the only reasonable way to plot things (with d3/protovis coming in a distant second) and it's supercharged by Hadley Wickham's work. It's also totally crippled for "general purpose" programming and turns to C every time you need more speed.

R is really good for playing around with data, but if you need to do some number crunching, R is horribly slow. This slowness offsets the advantages of having statistics specific features. Python itself is not too much faster, but PyPy is. I would pick R in most cases, but when speed is a factor, another route may be better.