That's interesting, because PDFs are an inherently un-streamable format. It requires seeking within the file to work (hence why command line PDF tools can't take STDIN for the PDF file - or they fake it by copying to a temp file first).
Well, I imagine that they download the file in the background, and if the viewer seeks to a non-downloaded part of the file, then it just blocks with "Loading" until that part is downloaded.
I wonder how google got around that problem.