On modern multi-core CPUs, memory has become progressively slower relative to the CPU. Even without a cache miss, a precomputed lookup table in a vector occupies precious L1 cache that could be used for other work. For small functions like days-in-month, pure computation can beat a precomputed lookup table. Wójtowicz uses it alongside branch prediction to argue that performance intuition is always wrong by definition — architectures change, so always benchmark hypothesis-first.