That said, there are some problem domains where better optimization is the only way to differentiate. Eg, I'm a short span away from releasing version 0 of Well

That said, there are some problem domains where better optimization is the only way to differentiate. Eg, I'm a short span away from releasing version 0 of Wellposed's numerical linear algebra product/library, and this is a product where if your code isn't at least as fast and usable as the alternatives, no one cares. Thankfully My code seems to be >10x faster on the problems that matter, and trending to be substantially more extensible and usable than alternatives. If "optimization" of some sort isn't fundamental to your business, you don't have a real business or value prop for someone.