Just remember that these are not genes (or, more correctly - alleles) that determine, in an absolute manner, whether you'll get the cancer or not (unlike, say in the case of the mutant gene for Cystic Fibrosis). They are variants that, when present alone or in combination increase the risk. It's a bit like with insurance - when you're a twenty-something, living in a large city, and want to insure a sports car, you'll pay a greater premium than a middle aged small-town father insuring a minivan, because the risk that you'll have an accident is several times higher. But that doesn't mean that all urban twenty-somethings in sports car will crash (in fact, most of them won't), and it doesn't mean that the minivan driving fathers never crash. It't the same with the association of genetic variants with cancer - there is no causative relationship - there is only an increase in risk. Which means that the test can be informative (to a varying degree), but is never definitive (unless it's a simple Mendelian trait, like CF, which the common cancers are not). Read this informative post on The Evolution & Medicine Review for a sobering view on genetic association in complex traits. Also, if you follow the link to the company page you'll see (in the News section) that the main paper (in NEJM) reporting the association was a study done on the Swedish population, and it's by no means certain, that it'll hold for other population backgrounds. All this doesn't mean that it's worthless, just an advice to take it with a grain of salt.

Still, not much of a publication record for someone so long in the business. Seems he is just very good at getting government money and jumped on the bioterrorism bandwagon. And as others have said - he'll identify mostly amino acid sequences that do not form stable folds. Actually, most proteins are built using a relatively small set of blocks or folds (think Lego) - cf. the CATH database http://www.cathdb.info/, so finding sequences that do not occur naturally would not surprise me.