"Superoptimal" reduction strategies

[cwgoes]

Namely, those with the possibly of asymmetrically-better performance than Levy's optimal reduction.

As far as I can tell, the only candidates here are forms of memoization which attempt to detect syntactically identical structures during the reduction process which can then be linked and evaluated only once.

Hash consing may have the most prior research.

Space-time tradeoff is the main concern here, I think.