It is difficult to provide a general answer to these kind of questions; they should be looked at on a case-by-case basis. Roughly spoken, the standard CHARMM charge optimization procedure requires breaking up the molecule into relevant fragments first, as shown in the CGenFF paper and tutorials and briefly explained
in the FAQ. If the high penalties (how high are they exactly?) are located at the interface between 2 fragments / groups, then it doesn't seem to be worth trying to optimize them. If they are within a group, then it might be worth optimizing that group (ie. all its charges).
I know it's very vague of me to say "relevant fragments", but this cannot easily be explained in a few sentences; that's exactly why I started with stating that it is difficult to provide a general answer.