I dearly hope something along these lines works, especially given the author has a repository and thinks the last set of optimisations may have been what introduced a bunch of the current bugs - if that's so, backing out those patches and relying on modern machine's performance to make the optimisations not matter might work out really nicely.

Yeah, I'm being hopeful from a complete lack of expertise here. But it'd still be -really- cool if it worked out.