I created induced dipole style polarization pair style for lammps. Unfortunately it can only work in serial because of memory and cpu time constraints and also the virial/pressure is most likely off. I mainly use it for simulations with a couple thousand atoms and box lengths of ~25 angstroms and it works well enough for my purposes.
The code is here: http://code.google.com/p/lammps-induced-dipole-polarization-pair-style/
Building it works normally, you might want to use make serial though. Examples are in /polarization and I also put a thesis whose 8th chapter talks about some of the implementation details in there. You probably want a coulombic cutoff of around 12 angstroms so the static electric field is fairly accurate. The polarization routines are copied from MPMC (also on google code) btw.
University of South Florida