I've tried every imaginable constellation, (SHAKE, NVT, NVE, Berendsen/temp, ...) always with the result that my water molecules stick around their original position. It's the same problem as Brian Novak (http://lammps.sandia.gov/threads/msg08733.html) had two years ago. Using lj/cut/coul/cut works fine, maybe there is a problem with the switching function?
Greetings Robert