- an optimization is considered lower priority than feature work and bugs, so never gets staffed
- it's considered risky even when it's not
- it's difficult to quantify the value until the works done
Aside from that, just because it may be more cost effective to use 1000 servers and 1 developer versus 200 servers and 3 developers, doesn't mean it's a sound choice. One invests in people using their brains, the other mostly disappears into raw material supply chains, building servers and infrastructure and burning fuel to run them.
They have a very hard problem on their hands of estimating the revenue growth new features will bring, and compare them to costs savings from better code. I don't think anybody here has enough data to decide on this - it's not impossible that nobody there has enough data either.