I think when you need more precision

There is no straightforward answer to this question. A memetic algorithm is just an extension of a genetic algorithm. It uses local search to reduce the likelihood of premature convergence. But according to the No Free Lunch theorem, there is no single best optimization algorithm, so this doesn't necessarily mean it is better. The best way is to check papers with problems similar to the ones you want to solve and then pick the one that performs better. But I wouldn't limit myself to only these two algorithms. Take the one that is best suited for the given problem. Here are some papers comparing a Memetic and a Genetic algorithm:

https://link.springer.com/article/10.1007/s43069-021-00087-8

https://ieeexplore.ieee.org/document/6921879

https://arxiv.org/abs/1004.0574?context=cs.NE