> 
> For large <for-each not preceding> would it be quicker to do
> sort list
> when current-node not equal to immediately-preceding-sibling then DO
> otherwise look for the next following-sibling not equal to 
> current-node

Yes, this would almost certainly be faster.
> ?
> 
> I don't know if internally Saxon already does something 
> similar to this. Or is finding immediate siblings costly?
> 

If you use <xsl:for-each-group group-by=expression>, Saxon will build a hash
table to group the nodes, which is probably more efficient than anything you
could achieve by hand.

Michael Kay