I think it could fit even more domains using a Bloom Filter or similar probabilistic data structure. With a chance of false-positives of course. But that's a trade-off the project already makes.