The reasons are the same as for HTML parsing really. Your regular expression will do what you want it to do, probably, and then you'll come to trust it and it will come back to bite you when you meet an unexpected case.

Does it:

• ignore code which is commented out?
• allow for attribute order changing?
• cope with the characters < and > appearing inside attributes, or CDATA sections?

There are probably a hundred more things you'd have to think of to make your regular expression solution bullet-proof, by which time you might as well have written your own XML parser.