Our species produces a substantial proportion of individuals with disabling and seemingly maladaptive traits. Natural selection, the ruthless elimination of unwanted infants and genocidal cleansing have not changed this. The heroic model of evolution as a crucible that burns away the dross of humanity and purifies the race has been tested thoroughly and refuted. Evolution doesn't work that way.
It is possible to argue that some high-functioning but apparently maladaptive traits might have contributed to group fitness in the heroic sense (see, for example, Nettle 2005; Spikins 2009; Hagen 2011; Nesse 2011; Nettle and Bateson 2012), but hard to explain the emergence of challenging behaviours and chronic dependence among primates that did not already possess advanced cognitive skills and high levels of altruism. Organic evolution remains the simplest natural explanation for the empirical evidence of contemporary anthropology, but the modern synthesis is frankly implausible. It requires us to argue that our ancestors became better equipped for survival and reproduction than more conventional chimpanzees because they produced hairless, helpless, congenitally macrocephalous infants.
In human populations, genetic evolution seems to be an equilibrium-seeking process that creates resilient complexes of checks and balances that protect deleterious genes from selective winnowing. One of the earliest illustrations of this is the textbook example of sickle-cell anaemia. The sickle trait is autosomal recessive, and is only expressed when the infant receives a copy of the allele from both parents. When expressed, sickle cell anaemia is very debilitating, and shortens life expectancy significantly. Individuals that carry one copy of the sickle gene and one normal gene are generally healthy and better able to resist malaria than those with two copies of the normal gene (Allison 1954). In populations where the allele is rare, the sickling trait is unlikely to be expressed and so will be able to hide from natural selection.
Another textbook example of inheritance patterns that seems to buffer populations from irreversible change is polymorphism. In the snail Cepaea nemoralis, for example, colour and banding patterns are governed by a handful of alleles. Different colour and banding morphs are better able to avoid predation on different types of vegetation, but the whole spectrum of traits can be reconstructed from the genes of survivors even in situations where some forms have been eliminated by predators. This means that even when environments change quite substantially from generation to generation, the snail population survives by re-assembling all the critical morphs.
These well-documented examples suggest that some of the 'fittest' genes are those that swim in large gene pools buffered against irreversible change by laws of large numbers. They hide behind dominance relations or become incorporated into stable polymorphisms. In situations where they are expressed, the fittest genes are those that code for some benign trait that does not undermine the carrier's viability or destabilise the attractor that sustains it. Possibly the fittest genes of all are those that code for nothing or mitigate the destabilising effects of deleterious genes. They may slip into the 'junk' DNA of a really successful host species or hitch a ride across species barriers with a virus - ideally one that never kills its host. Fit genes certainly do not go head-to-head with other alleles in a competition that only allows the strongest to win through. Genes maintain their fitness by hiding from natural selection.