Once you get results from CAESAR II you can review those results to see how the model responds. If the node moves: Is the restraint resultant friction force equal to mu*N; Does the friction force vector line up with the displacement vector? If the node does not move, is the restraint resultant friction force less than mu*N.
Are your results sensible?
Quite a few years ago, we had a low value for friction stiffness. Jobs converged quicker (important with the PCs of the day) but lower stiffness allowed supports to slide before they actually would and that, then, walked the model a little farther from reality. Based upon those observed model results, we increased the default stiffness.
Friction is a very "sticky" subject. We include it in CAESAR II to produce conservative results, maybe very conservative results. I, personally, do not expect an accurate prediction of system response. A much more sophisticated analysis algorithm would be required to attempt a better solution.
I usually try not to fool with friction parameters to find convergence. Instead, I work with nonconverging nonlinear restraints (+Ys or Guides with gaps) by making them linear (Ys or Guides with no gap, respectively). It's easier to evaluate the output for these points.