The sfImageMagickAdapater::loadFile() method is ignoring the “extract=>x” parameter and running the identify command for all pages of a multi-page PDF (i.e. "identify foo.pdf" instead of "identify foo.pdf"). In many cases all of the pages of a PDF will return similar results (i.e. width and height are often consistent in multi-page pdfs), but in some cases this could return incorrect data. In addition, adding the page specifier to the identify command has corrected errors in the "identify" return value for several PDFs that failed for me when not using the specifier.
I've attached a patch file that addresses this bug.