You're forgetting that DNA is far from the only input. There's also environment. And you're underestimating the chaotic, nonlinear system that is the material world. We can't accurately predict the weather a month in advance, what makes you think you can accurately predict the life of a human being?

Yeah sure you might find low hanging fruit and get free increases in "performance" by 10%-30% and bugfixes by simply optimizing the turing machine without any tradeoffs but most of the value is in optimizing the programs and data that are available to the machine. The best part: It doesn't require any modification to the turing machine, the effects appear very quickly within roughly two decades after birth and very often they can be upgraded later which means we don't have to deal with obsolete or buggy turing machines and even if they installed a suboptimal program and can't or don't want to remove it then next generation will still install new programs from a clean slate which is still more desirable than turing machine optimization that requires multiple generations to take effect. Heck by the time the turing machine optimization becomes available it might not even be optimal anymore because it was based around our needs X generations ago instead of what we need now.