> I want to strike up conversation/thread about the xxx:node-set, if
> anyone interested.  I am very frustrated with the lack of consistent
> performance I have witnessed using the xxx:node-set function with
> MSXML4 Parser.
> 

You haven't shown us any code or any performance measurements, so it's
difficult to make any comments. If you provided sample code and measurement
data then people could see how other products compared, or tell you how to
improve your code.

There are at least two possible implementation strategies for result tree
fragments and xx:node-set() in XSLT 1.0: one is to build the RTF as a
regular tree, in which case xx:node-set() essentially costs nothing; the
other is to use some other internal structure for an RTF, in which case
xx:node-set() might be quite expensive. I've no idea which strategy MSXML
uses, but your measurements might help to shed some light, if they are
carefully made.

Intrinsically, building trees with any product is likely to be an expensive
operation. That's one reason that XSLT 2.0 sequences are so useful, because
they involve less overhead.

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