Highly multivariate random walks are excellent at finding local optima, and poor at descending from those hills to find a higher one.
If something requires, say, five individually-deleterious mutations, each of low probability, it's quite plausible that these will never align in a single organism.
A localized entity responsible for the inefficiency doesn't sound like a local minimum, though. It sound like a place where you should (eventually through mutation) find the more efficient path.
There may be a more esoteric explanation... Imagine higher efficiency has a price : plant lives a shorter life, and reproduces less. Then you want to fine tune for the right amount of efficiency in your environment. Then it's advantageous to have a single efficiency lever, instead of a dozen systems to fine tune jointly... Just a guess - I know nothing about bio. :)
That is approximately why C4 photosynthesis hasn't swept the field and replaced C3 entirely: it's an advantage in high-light, dry conditions, whereas in wet, dark conditions, the more efficient use of ATP in C3 fixation provides a comparative advantage.
FWIW I'm quite bullish on the type of research in the Fine Article; humans have the ability to provide choice conditions to our crops, and so we can maximize yield without (much) concern for the tradeoffs experienced in the state of nature.
I'd predict, for example, that we'll have a C4 rice crop out of the lab and in the field in ten years or less.
If something requires, say, five individually-deleterious mutations, each of low probability, it's quite plausible that these will never align in a single organism.