> Maybe the question should be reduced to:
> 
> "What information is arguably lost in the identity transform?"
> 
...
> 
> Within the restricted world-view of XSLT, of course (which we 
> frequently 
> adopt happily) the answer is "nothing", insofar as 
> "information" is defined 
> as "that which has been expressed in an XML information set 
> (as implemented 
> by a conformant XSLT processor). An XSLT processor knows nothing of 
> anything else.

It's actually not quite "nothing".

- unparsed entities make it into the XSLT input but not into its output

- base URI is lost

- the isID and isIDREF properties of elements and attributes are lost (I
think - need to check)

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