I have tried it, and the only thing that jumps out at me is the ability
to use regular expressions to break apart badly marked up data.
Otherwise, most of what I do can be well handled by XSLT 1.

XSLT 2.0 has lot of other features apart from regular expressions. For
e.g. for-each-group (a newcomer now doesn't need to learn a
complicated Muenchian grouping algorithm. it's also very scalable),
rich data typing (tightly integrated with XML Schema, which has many
benifits), a very huge function library, and quite a few more
benifits.

Saxon I can only make work with a bunch of annoying Java
classpath voodoo. That's enough extra pain for using XSLT 2 for simple
jobs, that I'm far less inclined to do so.

Using Saxon just requires setting saxon8.jar in the classpath. Xalan-J
requires setting xalan.jar. So complexity of using Saxon is same as
Xalan.

--
Regards,
Mukul Gandhi