Just to be clear, the first two of these things are (relatively) easy because the libraries that perform the functions are already written. Actually writing an XML parser is unnecessarily of tricky because it requires schema understanding (in <foo><bar/></foo>... is there a text element between foo and bar?). Writing a validator is tricky because validation is a hard problem.

That does not help at all: if you download two Trojan versions, with or without noise, you get a Trojan version with your method. Also, it would not work with an even vaguely non-braindamaged watermarking scheme (the noise might be a key with which the rest of the info is encrypted... your method yields a random number).