Looking at the geomean benchmark suite everyone uses, it's 1.33 for Svelte 4 and 1.67 for React. That's a mere 0.2x difference and most of that is due to row swapping performance from React iterating lists of thousands of elements -- hardly typical in most apps.
Your team will make THOUSANDS of decisions in your app that will make WAY more difference than 20%. Incorrectly nest some loops in your data processors and the performance difference will almost instantly outshine the update differences.
And all of this is without mentioning that the React team is working on a compiler that will offer many of the benefits you get from the custom Svelte compiler (though TBH, I dislike needing a special compiler for custom JS).
EDIT: I'd also add that the Stage 2 record/tuple proposal will make comparing props many times cheaper and should give a massive performance boost to React as this comparison is one of the things that slows it down.