hjalsberg has explained this in some interviews. roughly 3x speed up from going native, another 3-4x speed up from being able to actually do effective multi threading