I suppose. Some problems with mmap() are a bit hard to fix from user land though. You will hit contention on locks inside the kernel (mmap_sem) if the database does concurrent high throughput mmap()/unmap(). I don't follow linux kernel development closely to know if this has been improved recently, but it was easy to reproduce it 4-5 years ago.