Yes, obviously, because I didn’t share your experience, I must not have tried anything non-trivial.

I’m not sure what the extra CPU consumed by XML formats over binary data has to do with XSLT being expressive.

FWIW, all documents in word processor formats are impossibly bloated, and casting them as XML makes them only more so; I wouldn’t want to try to get anything out of them, and that’s just the same no matter if I was processing them with XSLT or Perl. OTOH, I’ve processed some pretty complex SVG and XSL-FO documents, and it wasn’t particularly unpleasant.

I wouldn’t do complex SQL-type joins in Perl; not sure why I should start doing them in XSLT. (Not sure why I’d be sticking purely tabular data in XML anyway; that’s not what it’s for.) For simple correlation of bits from the document, I’ve not had any particular trouble. But it’s hard to know just what kind of data and operation you’re talking about here without any examples.

Makeshifts last the longest.