I think there's a lot of potential around improving Unicode implementation and properties, adopting perfect hashing in places (and alternate, non-linked-list hash implementations for small hashes with known sizes), and finding a better concurrency mechanism.