And then the table structure, relations, indexes, and keys all get changed because it wasn't carefully designed the first time. Then changes propagate into a bunch of queries and code and it's a much bigger deal than doing it better the first time ;-)
My view is that the schema + queries is the essential performance consideration. The code that goes from one persisted state to another is largely plumbing implementation detail. Of course there also lots of precise logic there but it's a healthy alternate view. The first thing I do on any project I join is make a big ERD to know the names, cardinalities, and relations (and wonder how they keep changing that without an up-to-date picture).
A proactive solution of hiring a DBA to build things along with the FE/BE developers will likely never get traction due to costs. Businesses regularly put off performance and optimization until later and they still make good money in spite of that.