How close would this be to native performance? Will it use any hardware features to reduce the emulation overhead if both the host and the guest use the same architecture?