Please include the list too (none of us are as smart as all of us). I tried ewald/disp and pppm/disp on a smaller version of the input script /bench/in.lj and they seem fine when compared to using a large cutoff with lj/cut. I can’t see anything wrong, they all seem to be converging to the same value. Are you using a tail correction for your cutoff? You could also try tightening the kspace tolerance to make sure your pppm/disp is converged.
Stan