xhtml2html is a utility that does what it says: it takes HTML written in the XML syntax (specifically, in the limited profile of XML defined by q4c12-xml-desc) and outputs HTML written in the HTML syntax.
It does not know about XHTML's entities. Pre-process the input to replace them with their equivalent text.
Because the XML syntax can be a lot more verbose than the HTML syntax, especially with dumb encoders (like mine). Specifically, q4c12-xml-desc will generate XML that looks something like this:
<n1:html xmlns:n1="http://www.w3.org/1999/xhtml"> <n1:head xmlns:n1="http://www.w3.org/1999/xhtml"> <n1:title xmlns:n1="http://www.w3.org/1999/xhtml">Hello, world!</n1:title> ...
This is perfectly cromulent and it makes the encoder's job a lot simpler. However, it's obviously very wordy.
Compression helps, but it's not quite a silver bullet. gzip's maximum window size of 32 KiB also means that, for moderate to large documents, any extra repetetive verbiage is taking away space in the dictionary from more interesting things.