> It actually makes it hard to give general advice: On the 
> whole I assume
> (when writing answers, if not in my own code) that there is no
> optimisation going on.

Yes, that's reasonable advice.

To be honest, the main reason for this optimization was for XQuery, where
(a) there isn't an equivalent to xsl:key, and (b) there is a strong user
expectation based on years of SQL culture that the first optimization every
processor will attempt is a join optimization. But it's a nice bonus that it
works with XSLT too, except for those who earn a living by tuning other
people's code...

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