If want to make the calls by your self, you could likely calculate the alternate allele fraction for known sites intersecting with dbSNP, 1K genomes, ESP project etc and create a cut-off based on that.

I don't know if < 400 counts as ultra-high any more, but most variant calling software will take other things into account such as the quality distributions of the reference bases compared to those of the alternate, strand bias, etc.

That coverage is low enough that if you get a smallish number of candidates, you can load the bam(s) up in IGV and have a look. That usually helps to determine if there are alignment issues in that region or if things look off.