Just two thoughts as a "for what it is worth":

In the 1980's, bushingless chain was introduced for bicycles. Essentially all modern geared bike chains are of this design. Bushingless chain shows higher friction losses in the loaded direction, and lower loss when side loaded. When people talk about chain efficiency being known for a century, I believe a variable was introduced in the 80's that modifies that conventional thought.

The Spicer write up linked, and a different Human Power article on efficiency of different shifting methods, all include a feature that the bicycle does not have. On page 5 of the link, the write up specifies that the output shaft could be moved to set distance, in a way that I read as longitudinally as well as laterally. This allows each chain ring to be set at its optimal center distance. A bicycle does not move center distance as it is used. Deviations from optimal CD have a direct impact on losses from chordal action.

I may be misreading. Just thoughts.