> Animals are pretty inefficient at converting plants to calories, e.g. see ...
That is not what that graph shows. That graph purports to count all energy 'inputs' into the system, including human labor. This is not about 'how many calories did the cow / pig eat, versus what it produced'.
Here's an example.
Consider wheat grass. Both a cow and a human can eat this grass. However, the human can eat the seed, the cow / pig / goat can eat the entire plant. If you burn the entire wheat plant, and count the calories released you will get some number. Let's call it Y. If you count the calories in the seed that we eat, it'll necessarily be less, let's say X.
Ideally, if we're going to produce Y calories in the grain, humans ought to be able to consume as close to Y calories as possible in order to get the total benefit. Since humans can only eat X, and X < Y. Then it makes sense to feed Y-X calories to a low maintenance animal (so not beef), and then eat the animal. This will yield some number N < (Y-X) that humans can consume. This also provides a complete protein, fats, and other nutrients not found in grain (B12, which is not found in plants period, for example, thus preventing yet another synthesis process).
X + N < Y, but it is much closer to the actual amount of energy in the grain plant than just X. And then we'd need to count the energy expenditure getting the B12 and Vitamin D produced and distributed as well. And then the supply chain risk costs. All these things have costs that never get considered.