You might also consider excluding the regions with poor geometry from the NCS groups, to see if they are in fact different. You are also forcing NCS regardless of the RMS distance. I would refine a few rounds without NCS and with the default refinement parameters, and then then turn the NCS back on. Look in the .geo file for excessive distances, and examine these regions in Coot using the NCS ghosts and maps to decide if they should be excluded from the NCS groups.

Thanks Kendall, it is very good point. We should probably do something like this to automatically adjust NCS restraints during refinement.

Pavel.