I feel so certain that I messed up my min-cost analysis.

And I am 100% sure that faster is possible, and will get hit. I just can't be bothered to hand optimize a 1000-instruction tape.