I'm using the Muenchian Method to grouping my XML file. But I have a big
problem: my XML file contains cca. 6000 nodes, so the transformation can
be very slowly.
How can I make it faster?

It depends on a lot of things
- the XSLT processor you are using
- the hardware you run on
- your definition of "slow" and "not slow" (aka "fast")
- your options on changing any of the above
- your input
- the actual code you use
- whether the slow performance is really caused by grouping and
  not by the sheer size of the input or other things

suggestions
- use a faster machine
- check whether your grouping code uses keys (it should)
- get a profiler and establish beyond doubt that it's the
  grouping which slows down you transformation
- use another XSLT processor
- use an XSLT processor written in C, like libxslt or msxml
- check whether your XSLT processor has known performance
  problems with functionality which is important for grouping,
  in particular with the generate-id() function
- tune your environment, in particular the JVM if you have
  to use a java based XSLT processor
- change your input XML so that it is already grouped