In 1990, John Lamping published a paper proposing an optimal implementation of the untyped lambda calculus. Since that paper is 25 years old, I wonder how much we have advanced since. Thus, my question is: what is a simple description of John's optimal lambda calculus evaluating algorithm (or, in case we made improvements since, of the improved algorithm), preferably explained briefly on Haskellish-pseudocode?
Update: as I've learned more since I asked, I believe a valid answer could be simply a pseudocode for an unbloated algorithm that 1. maps pure untyped lambda terms to interaction nets; 2. reduces those nets and 3. maps back from nets to lambda terms such as that the whole process normalizes the initial lambda term optimally.