Well, that's good, but I'm still missing other features I get with Boost.Python in minimal code (typically ~3 lines of C++ per typical method (with named args and a docstring); add ~10 extra for more involved methods including virtual methods, callback wrapping, and PyObject-as-std::iostream translations). Other features are custom getter/setter functions for properties (boost::variant bindings with named accessors which return None if the value for name isn't set and can be assigned to), protected class access (we allow inheriting from a a C++ class in Python and calling it from C++ again), and the extract class mentioned above (boost::any bindings). Not sure how well SIP handles using objects via a shared_ptr (or similar class) and never by instance either, but this is likely hackable with some work.
With which my main gripe about is the lack of PyPy support. This isn't me just ripping on other bindings systems for the sake of it; these features are just required for complete bindings and that is more important than PyPy support (as nice as it would be).
"SIP does not support access to either private or protected instance variables."