Matts++ (like he needs any more xp but it's the thought that counts)

I asked this same question on the mailing list a while ago :)
If you still like the thought of XPath over DOM (as I did/do) but would like to preserve/edit/add... CDATA then I found XML::LibXML a great alternative to XML::XPath. (plus it works well with XML::LibXSLT which I beleive is faster than sablotron)

<edit> added some code </edit>

#!/usr/bin/perl -w
use strict;

use XML::LibXML;

undef $/;

my $XML=<DATA>;

my $parser = XML::LibXML->new();
my $xmlp = $parser->parse_string($XML);

print $xmlp->toString; #before munge

foreach my $node ($xmlp->findnodes('/foo/bar/text')) { 
    my $data = your_munge_function($node->findvalue('text()'));
    $node->findnodes('text()')->get_node(1)->setData($data);
}

print $xmlp->toString; #after munge

# functions
sub your_munge_function
  { return "munged $_[0]"; }

# data
__DATA__
<foo>
<bar>
<text id="text1"><![CDATA[La dee da de da.<br>Foo bar baz]]></text>
<text id="text2">normal text</text>
</bar>
</foo>
[download]