Yes, I use PyMol quite a bit. The source code is hideous, and the author likes it that way. It would take an immense effort to smoothly incorporate PyMol into a larger system like SciPy or Sage; it would be very useful if that happened, but I don't think it will any time soon.
I haven't used Mathematica 7 yet, but it looks like the tools it implements in one small library are decent substitutes for what's currently implemented in several big hairy bioinformatics toolchains. Look at how we operate now -- NCBI offers a bunch of free standalone tools; BioPerl and BioPython do some good glue, and then there are a bunch of visualization tools, and we FTP copies of whatever databases we need and keep updated manually. The extensible tools each have their own extension language, so a RasMol script won't load in PyMol (the tweaks, not the PDB file). Mathematica 7 appears to offer the basic components of this in one clean, well-organized standard library, controlled by one powerful language.
All that's great, but the especially interesting thing is their claim that the design of the whole system made all of this easy. The image-processing tricks and the unassuming Parallelize function seem like very specialized features, but they cobbled all of this together in, I guess, 2 years and some change, on top of a codebase that's almost 20 years old. It's like the anti-PHP -- brilliant design decisions all along that make adding new features easier, not harder. I have a feeling the Mathematica source code contains some very interesting ideas, and if it were open-source, it would be relatively easy to add additional features that seem like a big deal in existing tools.
I haven't used Mathematica 9.0 and its chemistry-related functions, but from the presentations, they look like practically useless toys for dabbles.
Can Mathematica do raytracing, display secondary structure, semi-transparent molecular surface and allow you to select coloring patterns? Is it easy to predict protonation states of molecules at different pH levels and electrostatic potentials? Can it do molecular dynamics?
Who cares that they have built-in access to the melting point of caffeine?!
I think the point is that you have to type about 30 characters to get a 3d plot of the molecule of choice. Unlike M-, PyMOL doesn't have the data built-in, so you have to find, download, load etc yourself.
Try PyMOL - the most widely used molecular graphics software, probably.