Re: speed optimization

  Well, maybe Saxon's reports are more fine-grained. chaning libxslt
is of course doable, but I have plenty of stuff on my plate already
and so far nobody requested something finer-grain.

I suppose if a huge template is identified as a bottleneck,
one can split it into several named templates. Named
template invocation and passing some former variables as
parameters wouldn't add too much overhead, wouldn't it?