> The code runs fine as is. But the processing time explodes (from 6s to
> 82m) if I change line 33 such that the select statement uses 
> a predicate on the node set. 

I don't think there is any intrinsic reason why that should perform badly;
but unfortunately writing your code to achieve good performance is
inevitably dependent (to some degree) on the processor you are using. So
this is really a product-specific question.

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