In regards to vcfutil.pl varfilter -D I also asked a question about this:

How to interpret -D file in samtools variant calling pipeline?

The thing to be clear about is that any position that exceeds the coverage of '-D' will be excluded from the results. When i run variant detection i usually set '-D' to be 3-5x the mean depth of coverage for the genome. This seems to work well, but still excludes areas where there is artificially high coverage.

Not sure what you mean about '-S'? For which, program?

As Ian answered I would set -D to be 3-5 times of the mean coverage. I usually keep it 3X. This way you would discard false postitive SNPs resulting in the CNV regions OR highly similiar regions like paralogs. -S is a p-value but I usually dont care a much about it and use a very low cut off .0001. There have been many papers those have used the same p-value for strand bias.

I wouldn't call 20x high coverage. I'd say high coverage is 100x, at least.