The limitations of XML cannot provide for roundtrip compatibility. Here is a possibly incomplete list of issues:

• XML 1.0 (apparently what browsers use for XHTML) only supports a subset of Unicode (see http://www.w3.org/TR/xml/#NT-Char). Some characters (e.g. the ASCII control codes) cannot be converted (or event quoted) and stripping those off seems to be the only viable option.
• HTML allows colons in tag names without attributing any special meaning (see http://www.w3.org/TR/2011/WD-html5-20110405/namespaces.html). A 1:1 translation to XML creates a namespaced element quite probably without the adjacent namespace declaration.