The key claim is that it has been specialized against the target architecture to tune each parameter optimally. But, yeah that was done 20 years ago and it is not clear how much the costs have shifted on modern architectures.
[1] https://nothings.org/computer/judy/ (2003)