The transfer of image data only happens when we yield inside the engine which only occurs when you hit a breakpoint in script. Otherwise we're just rendering to the OffscreenCanvas and letting the normal flow blit it to the screen (not doing any copying).

I just did some profiling on Firefox and I feel like the profiling result doesn't quite make sense but it's saying that the majority of time is spent calling Performance.now() from the clock calls in C++. I'm wondering if that's because we're calling it too many times and maybe we should just call it once per frame.