Then we’d have more cache misses and have to re-parse analysis from disk more, but we wouldn’t chew up unlimited RAM