This is the beginnings of a library to optimize PDF content streams. A "content stream" is the PDF term-of-art for a packaged sequence of drawing instructions; for instance, there is normally one content stream for each page in a PDF file. Many programs that generate PDFs do not take any care at all over the efficiency of their drawing operations, leading to documents that are much larger than they need to be, slower to render, and can cause renderers (especially actual printers, alas) to choke.
I say "beginnings" because the actual optimizer has not yet been written; I began this project in 2010 and almost immediately ran out of time to work on it. All that there is right now is an incomplete parser for content streams. Still, it could be useful as is. If you have an immediate need to make a PDF document more compact or efficient, modern versions of Ghostscript include a page optimizer:
$ gs -q -dBATCH -dNOPAUSE -dSAFER -sDEVICE=pdfwrite \
-dPDFSETTINGS=/printer -sOutputFile=out.pdf in.pdf
will probably do what you want.
All the present code is in Python, but the long-term intent is to write the optimizer in C or C++ so that it can be incorporated into QPDF, keeping a Python version around for reference. This is the reason for the unusual choice of license. It would also be nice to include an optimizer in common PDF creation libraries, eliminating the original problem at source, and so I will consider most any request for license adjustments necessary to incorporate this code into some other work of free software.