It's because that first invocation of sort is sorting the entire input (413MB), not just the unique words (less than a MB). The sort is probably O(NlogN), but t

It's because that first invocation of sort is sorting the entire input (413MB), not just the unique words (less than a MB). The sort is probably O(NlogN), but that's a big N. Counting by inserting into a hash table is much faster, at O(N).