When the true positive rate is very low, bloom filters can be much faster than expected because most queries will be rejected by the first couple bits read. In that case a blocked bloom filter can easily outperform a cuckoo filter since a negative result always requires two random memory lookups for the cuckoo filter. In exchange for memory usage Bloom filters can also be made faster for mostly negative query loads by running them under-full so the probability of any bit being falsely set is extremely low.
the fine details of the implementation can easily matter more than the overall data structure.
