In serialization section 4, phase 2b states that character maps are not applied
in elements included with cdata-section-elements.

This seems an odd provision, and I wonder what the rationale for it is? It
upsets the sequential pipeline of character processing, and it's not clear to me
that it has any user benefits. It encourages users to use cdata-section-elements
as a mechanism for making character maps apply selectively to different parts of
the result document, which will only cause frustration because it doesn't do
that job very effectively. Admittedly there are some use cases for character
maps - such as generating entity references in the output - that don't make
sense within CDATA sections; but the answer to that is for users to avoid mixing
the two facilities.

Michael Kay