While further research is certainly a good idea, I push back pretty hard on this example!
You have already violated a major assumption here, by accepting a non-standard-size transaction. Yet we’ve already hand-waved over the bin-packing problem, accepting that it can be suboptimal in theory while Good Enough in practice. If this is untrue, we have other problems.
Simplicity and predictability itself is a virtue, as long as the result is congruent in practice: miners will almost certainly agree with this, as we’ve seen them lose real rewards by trying to be tricky with block construction in the past! (I posit, far more than more optimal packing strategies will gain them in their lives).
I think if we were to analyze the difference in practice we will end up so far below the engineering salary required to implement and vet the more complex scheme 