At SCO in the mid-90s we were playing with very similar ideas to boost DB performance. The main motivation was the same then as it is now, don't block and avoid making system calls into the kernel once up and running. Don't recall if any of the work made it into product.