Saying rasterization or raytracing is slower or faster than the other is inherently flawed.
Some things raytracing will be way faster at, and viceversa.
When you need indirect contributions, or raytracing eyes, both things that happen in nearly every-single-shot out there in a creature show, bruteforce is already faster than bolting a raytracer on top of a reyes engine or spending a day computing point-clouds and storing 20GB worth of data per frame before you can even stand a chance to render the final look for that frame.
There are other things too, such as very dense meshes (sometimes unavoidable) falling into a single bucket. A raytracer, once the acceleration structure is built, will just run over it like bad roadkill, REYES will still grind to a halt on that bucket before it even starts thinking. Conversely, sparse cages with lots of disaplacement coming close to camera will always be faster in REYES rasterisation engines than when you have to compute all the new geo beforehand, store it, then raytrace it, and send the acceleration structure into chaos memory wise half the time.
It’s not quite as black and white as “raytracing is slower” these days in production. I don’t know why people keep harping on that string, but I suspect most never tried one of the good ones 

