On 704A - Thor, the policy-based hash table easily passes under the time limit, but the unordered set version TLEs on case 56.

With gp_hash_table: 181117819

With unordered_set: 181133673

Both submissions use a custom hash function, so it can't be due to collisions. On test case 56, the unordered_set suddenly gets TLE but it's only slightly slower than the gp_hash_table on all the other test cases. Why is this?