In improving my previous solution, I counted the cycles that each arm was idle (8/19 on arm 1/2) and did my best to reduce that, even if it meant kind of inefficient arm work; as long as a change prevented an arm from doing nothing, and what it was doing was vaguely productive it was a good change. As a result, this solution has (11/4) idle cycles, a reduction of 6 on average. The tape loop that results from this is 5 cycles shorter than the previous solution's, suggesting a pretty positive indication for my theory; I was able to convert arm idle cycles (in the abstract) into saved operating cycles with 83% efficiency. I suspect there's a degree of diminishing returns to this barring serious work-process changes, insofar as doing this isn't work-process changes in the first place.