[rurban]

See, everybody can see how you cheat your benchmarks. It's unrealistic to measure only the perfect hash function, when you discard the cost of the deserialization and false-positive checks.

That's always the case when dealing with you academics.

We dont care how you define your costs when we laugh about that. In reality you have to load the keys, the data structures and check for existance. You even discard the existance checks, by omitting the false-positive checks. Checking against a non-existing key will lead to a hit in your check. Only very rarely you know in advance if your key is in the set

[rurban]

Also, you miss the space costs for the ordering.