There was another project posted a few days ago that was quite similar and someone asked the same thing to the author. They said they tried that first and they

There was another project posted a few days ago that was quite similar and someone asked the same thing to the author. They said they tried that first and they got a 10x speed boost by doing it manually. It is the same reason why database engines write their own backing cache logic: the kernel's paging logic is generic and on-demand, while someone that's aware of the actual usage patterns can prefetch and pipeline the right data before it is used.