As others have those benches are also largely based on the underlying code of a page and how it was written -- eg engine-specific optimizations -- and finally browser optimizations, as pointed out above in the case of Edge which is slow AF., etc. so that's a poor measurement. Most accurately you'd bench rendering SVGs, AA'd fonts, and both compressed and uncompress raster images, and finally up-/down-scaling and filtering (NN vs 2xSAI, etc. -- if applicable, IDK what engines use). Benches would have to all be local content, obviously.
But render speed is definitely not the be-all and end-all. There's compatibility, too. It's no good having a lightning browser engine that can't render SVGs properly or isn't up to date with HTML 5.
That all said, I'm with
@CaffeineAddict on this: "Best" is a pretty relative term. For example, an engine could (IDK if any do use this method) protect against canvas-based fingerprinting by randomizing the presented canvas size and start coordinates, then internally up-scale it (using magnitudes that did not break layout or provide too specific of a range).
So for the casual user, that'd be a nuisance, but for me, that'd be points. Bonus if it would include special APIs to set various render methods and even throttle speed to make it harder for an adversary fingerprinter.
Thus, I'd base "best" on how much privacy it could offer before the browser tweaks come into play. Most casual users would be like, "IDGAF, I just want it to work OOTB." So "best" for them would be speed + accuracy + features... which is fine and totes the perogative.