I see nothing wrong with the code. PDFEdit is not presented with operators one at a time from the page stream. It is presented with an abstraction of the graphical elements. Operators that change the graphics state (such as cm) will update the graphics state presented to PDFEdit, and their effect can be seen by inspecting the graphics state. The simple example does not dump the graphics state. The graphical elements presented may be graphically equivalent to, rather than precisely, those found in the page stream.
The job of the example is to help understand what the calls do. It is not to provide a handy path dumper with accurate coordinates unless that happens to be a side effect.