That is, given "fully rational" reproducers, the gene pool would get increasingly better and produce better and better humans over time.
Environmental conditions change over time. What may once have been an advantageous trait in a human population may become deleterious in some form at a later time or different environment (i.e. by spatial migration).
Case in point: Central Africa is plagued with malaria, a disease caused by a parasite of the Plasmodium genus which infects red-blood cells (RBC). Sickle cell anemia (SCA) is a hereditary disease which can be deadly, unrelated in its cause and prognosis to the Plasmodium infection, BUT, it confers a resistance to Plasmodium infections due to the RBC of SCA-patients having an aberration in the shape of their RBC's. Here is a selective pressure caused by malaria that "promotes" the profileration of SCA in the local population. As a result, what you have is a large section of Central Africans that are Sickle Cell Anemics, whom have better survival rates than non-anemics in that environment. But if you take these anemics to northern Europe or North America, where malaria is virtually non-existent (spatial, environmental change), the anemics have a higher mortality rate than the rest of the local population.
You get what I'm saying?
Natural selection doesn't produce better and better humans over time - it produces humans that are better able to survive in that era, in that place.