What efficiency measures is a function of how broad or narrow one's view of a system's thermodynamic universe is.
A narrow view would be to look at just the thermal efficiency of an engine itself. That is, what percentage of the chemical energy contained in an engine’s fuel gets converted to power output on the engine’s driveshaft.
A mid-range view would consider efficiency as the ratio of the work done by one’s robot to the amount of fuel its engine consumes. In this case, one has to consider, not just the thermal efficiency of the engine itself, but also how efficiently that power is coupled to the work the robot is doing. This is the impedance matching problem from before.
At a global view, the system's thermodynamic universe becomes an economic one. That is, to the farmer efficiency is now the amount of profit generated per robot versus its cost of operation. In the end, this is the most important figure-of-merit to the farmer.
It would certainly be advantageous if one could break down the question of engine efficiency into three such well-defined categories. But such is not the case.
The first complication is that an engine's thermal efficiency is dependent on the power load and RPM it is running at. This effectively couples the local and mid-range views together. One can’t analyze an engine’s thermal efficiency without considering at the same time the mechanical system that engine is driving.
The second complication is that the laws of physics put hard limits on the maximum thermal efficiency any heat engine can attain. Ultimately what physics tells us is those higher efficiencies require higher operating pressures and temperatures. These lead to higher mechanical stresses on an engine’s design, leading to more frequent breakdowns and required maintenance. And it is in the area of maintenance costs where the local view of thermal efficiency crosses over into the global view of economic profitability.