I've run tests with the download window reduced to 1024 blocks (like BC) to see if this type of benefit can be achieved in such a small window, allowing blocks to be memory buffered. But in this configuration BN slows dramatically to essentially BC speed.