Comment by monocasa

What do you account for the perf win over Qemu? A bunch of micro optimizations, less abstraction layers, or something more systemic?