You’re talking about messiness in the genotype, but selection pressure happens to the phenotype. The latter I think controls more of what ends up in the former.
Which is what we see all the time in programming: software wins because of its effect, not usually the lines of code needed (or not) to get there.
I'm not sure I understand your argument, but if large portions of the genome have no effect on phenotype, then perhaps we can both agree that those regions would not be under selection. Pressure on the phenotype cannot directly alter the genotype (no Lamarkism), but it could select for or against the fraction of the population with a particular genotype. But only if the genotype produces a phenotype.
Wouldn't assuming a one-copy rule end state is natural cause immense issues with recessive genetic disorders (which could in turn select against it), and also make chromosomal-crossover produce sterilized but viable young at a much higher rate?
Which is what we see all the time in programming: software wins because of its effect, not usually the lines of code needed (or not) to get there.