Just in case this helps: it doesn't appear to be exactly the same data embedded into the .CR2 as that which can be exported as a stand-alone recipe (a .VRD file), but it does appear to be wrapped in exactly the same text string, namely:

CANON OPTIONAL DATA

After a check of another couple of hundred untouched files it appears that the camera never appends this string. Obviously, it's possible that this might come up inside a free-form text tag, but if it's possible to examine the portion of the file between the last declared block and the fsize idea of the file length (hmm, showing too much C there: sorry), then looking for those strings and keeping the data inbetween ought to be fairly safe. ie: only accept the string as a block delimiter outside of the TIFF structure.

HTH,
--
Ian.