Currently, about 15 standard library modules are supported, such as re, math, random, collections and getopt. More will certainly follow (e.g. we have a datetime implementation in SVN), but this is a slow process, and it will always be limited to a relatively small subset of the standard library.
However, I don't think this is a really big problem in many cases, because in principle you can always generate extension modules ('shedskin -e'). This way, you get a potentially massive speedup for one or more speed-critical parts, while still being able to use arbitrary Python modules and dynamic constructs in the 'main' program. And everything is still in pure Python :)
Note that I don't think Shed Skin is the ultimate solution to improve Python performance. It's a cool toy that may be useful in certain cases, but what we really need is a good JIT, something that may come out of the PyPy project.