I would use the IMAGE type, as opposed to any sort of TEXT. While I am not debating what Rickard Andersson said, XML is text, I can still remember an issue where an XML with French characters got its content ruined when read as text (as in XmlDocument.LoadXml), due to a combination of the XML processing instruction encoding declaration, the character set in the database (and maybe something else); basically the round trip didn't work (write/read).
With a binary storage, it is guaranteed to work, since your application completely controls the encoding/decoding of the binary stream. You can also employ compression much easily, a very much needed feature if your XML files are, or will ever grow, large.
The only downside has to do with the ability to "peek" more easily at the XML (i.e. tools will not attempt to show you text, but hex bytes). Otherwise, full-text indexes would work just fine...