XML::Simple's name is misleading, it sounds like "a simple module for complete XML handling", but it's not - it's more like "simplistic XML handling for simple XML". I find it's great for reading simple XML config files that have been designed to work with XML::Simple, but it is not an all-purpose solution, and in your case is very likely not appropriate because, among several other things, it very often doesn't maintain an XML document's structure when reading a file and writing it back.
Anyway, I'm inclined to agree with the other monks' suggestions for XML::Twig, which is great when you want to process a file piece by piece. For the case you describe, if the file isn't so big that loading it into memory is too expensive, then XML::LibXML is fine too. For example, the following deletes the <bar> element if its child <quz>'s text content is "baz".
use warnings;
use strict;
use XML::LibXML;
my $dom = XML::LibXML->load_xml(IO=>\*DATA);
print "### Before:\n", $dom->toString;
for my $el ($dom->findnodes('/foo/bar/quz')) {
$el->textContent eq 'baz'
and $el->parentNode->unbindNode;
}
print "### After:\n", $dom->toString;
__DATA__
<foo>
<bar>
<quz>baz</quz>
</bar>
<x/>
<bar>
<quz>abc</quz>
<y/>
</bar>
</foo>
Output:
### Before:
<?xml version="1.0"?>
<foo>
<bar>
<quz>baz</quz>
</bar>
<x/>
<bar>
<quz>abc</quz>
<y/>
</bar>
</foo>
### After:
<?xml version="1.0"?>
<foo>
<x/>
<bar>
<quz>abc</quz>
<y/>
</bar>
</foo>