This also seems to misunderstand evolution. You can think of evolution as a hill climbing optimization algorithm, with a complicated and constantly changing fitness metric. While it is true that any given change is random, and even 'good' changes still only succeed with some probability, over a long time period we see a series of incremental improvements. If there is an opportunity to increase efficiency it is likely that evolution would have found it, given how long the process has been running. Of course, this become less likely when we are discussing new features (the thumb is less optimized than DNA because it is so recent), and as the benefit of these optimizations becomes less. They also become less likely if their are fewer evolutionary paths that might lead to them.