Since posting this I checked the speed comparison page where buzhug is shown to perform well against some lesser known databases and poorly against SQLight. Now any new code base is going to start off slow but it begs a question. If you are aiming to serve the same niche as SQLight then why not simply develop a pure Python API for the SQLight database and get the best of both worlds? That way you could keep the Python abstraction clean and take advantage of a fast and very well tested database engine.
If this sort of thing interests you, take a look at PyTables http://www.pytables.org/moin - it uses similar python constructions instead of SQL. For some kinds of things it beats SQLite. Only downside is it requires the HDF5 libraries.
