> I deal with geographic data which can be very large, and in 
> many cases must be dealt with as a whole in order to deal 
> with spatial relationships.  I use xslt to process it 
> (elegantly, capably, correctly), but it seems a shame to 
> devote that much memory to a file for so long and only be 
> able to use 1 core on a multicore machine.

For this kind of use case the simplest way forward might be to do parsing,
transformation, and serialization in separate threads. Potentially of course
each step in a pipeline can be done in a separate thread, with very low
coordination overheads. It's probably quite possible to configure Saxon to
work that way today.

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