Thanks David for info about  . But webdesigners don't like to see
just a space not knowing exactly what that space is. I will let the
protection of character entities be optional in my solution.

Your test document:

<!DOCTYPE x [
<!ENTITY x "x]>">
]>
<x>&x;</x>

is wonderful. I think I have fixed the problem with this REGEX finding
DOCTYPES having internal subsets:
regex="&lt;!DOCTYPE.*?[&quot;|&apos;]\s*&gt;\s*\]\s*&gt;" flags="s"

CDATA problem is also fixed, new version of stylesheet will be oploaded soon.

Yes I know the power of xsl:character-map, I am already using one in
the solution but I will rethink the problem and also consider making
own entities with a DOCTYPE in top of stylesheet.

I will not bother the list with more problems, but will use the coming
days getting the test suites going.

Thanks for the help.
Jesper