Think about it, the human genome already contains genetic encoding of keratin, it wouldn't have to evolve (incrementally bruteforce) a full protein code to "protect and repair" damaged teeth. It would just need to "happen onto" accidentally expressing it somewhere in the mouth: perhaps the mucous membrane lining the inside cheeks, perhaps the tongue, perhaps some glands in the mouth. Accidentally expressing a gene in a cell type that didn't before is much easier to occur (i.e. more likely) than generating a new functional protein: all it takes is a change in the binding site (or promoter region) so that the relevant cell type (say lining the mouth) would express it, conditionally or unconditionally.

If this were effective, our bodies would probably be doing it already.

Just to clarify: even if 2 people had the exact same genetic coding for proteins, but different coding of promoter regions, then these will have different binding affinities, modulating when proteins will or wont be expressed and at what rate. So when considering a population's genome statistics, there is already a spectrum of promotor region codes in the population, if this keratin presence on teeth had significant advantage, selection pressure would already have increased that level towards optimum.

The only caveat for my reasoning would be if it were discovered that this is exactly what happens in a healthy mouth, and that we recently discovered that conventional toothpastes have been stripping such layer of keratin by abrasion.