I think this could make linear scan take up to 20% less time, which would lead to maybe a 5% improvement on -O1 compile times.