I'm a bit stumped by the performance issues we're having.  We have an
xslt 1 compliant stylesheet that takes 10 minutes to run in xslt 1
processors (Saxon, Xalan, libxslt) and a few seconds in Saxon 9B.  I'm
not sure what it can possibly be.

I've found that processors may behave very differently with respect to
coding patterns, or algorithms. For example, some processors cope well
with sibling recursion and others don't.

Some processors may do intelligent high-level optimization, others may
not be built to discover such possibilities.

When doing some random tests, I've seen Saxon to scale best and perform
most evenly with regard to input size and coding patterns.

With xsltproc, --profile may yield some clues. Other processors have
similar switches.