Slightly off-topic but i think a key to successful large scale Schematron is to not run the XSLT direct from the command line.
I have had success on projects using xproc and with c# and java servers that call the xlst directly, avoiding eg jvm startup time and allowing precompiled java easier.
You can still have batch files or shell scripts but these call the schematron through a rest service.
On the other hand, i suppose now we are all in the cloud we dont need to worry about performance, isnt it?
:_)
Rick