> >A prefix used in an XPath doesn't stop the XML from being namespace 
> >well-formed, so I don't see why that namespace needs to be 
> declared in 
> >that particular file. It seems an artificial restriction.
> 
> Not to me.  Importation and inclusion doesn't happen at a 
> syntax level, it happens at a semantic level. 

I'd call it a design choice, rather than an artificial restriction. 

It could have been done differently. We chose to depart from relying
entirely on the XML-level namespace bindings in the case of
xpath-default-namespace. Doing it differently would have had some negatives,
like <a:element/> not meaning the same as <xs:element name="a:element"/>.

Michael Kay
http://www.saxonica.com/