The guy has been dead for ~5000 years, but they're looking at SNPs linked with these phenotypes today; surely over 5000 years not all of those associations remain.
I would consider that a strong indication at least that the authors aren't talking completely out their asses.
I think what you might be trying to say is (1) the presence of a SNP may not be sufficient for the occurrence of a phenotype, or (2) another SNP in high LD may be the true causal SNP.
Imagine the kind of inaccurate results you'd get by comparing two versions of your code with GWAS instead of reading the commit history.
Knock-Out studies are all about determining the consequences of a genome change.