Consider:

      ⎕XML'<p class=myclass>hello world</p>'
DOMAIN ERROR: Quote expected at start of attribute value at character 10 (line 1, column 10)

or:

      ⎕XML'<div>hello<br>world</div>'
DOMAIN ERROR: Tag mismatch at end of text; expected </br>

or even:

      ⎕XML'<div /><p>hello world</p>'
0  div                     1
0  p    hello world        5

This last one is a trick question.

And consider a million other things the browser will handle, but ⎕XML won't handle or will handle differently.

What to do?

Once we are committed to carting around the 200 megabyte HTMLRenderer, we can just hand off converting HTML to XHTML to the browser, yielding a suitable argument to ⎕XML:

GetXML←{
     j←'new XMLSerializer().serializeToString(document.getElementById(''',⍵,'''))'
     ⍺ ExecuteJavaScriptSync j
 }

The browser is a mighty thing, and it's now as integral to the APL interpreter as anything.