Sure: If you can optimize the software that runs on many millions of computers then you can have a huge impact. If you run those computers yourself you can even save money.
But the vast majority of software is one-off stuff. It makes no sense to optimize it for performance instead of features, development time, correctness, ease of use, etc.
> But the vast majority of software is one-off stuff.
Is it now? I can't think of any example, except a few tools we use internally in a project. Everything else I use, or I see anyone else using, has an userbase of many thousands to millions, and a lot of that is used in context of work - which means a good chunk of the userbase is sitting in front of that software day in, day out.
It'd be interesting to know, say, what percentage of software developers work on programs that have less than a million users.
I wouldn't be at all surprised if it's the majority. For every big-ticket software offering there's going to at least be things the user never interacts with, like a build system and a test suite and a bug tracking system and whatever else. And there's just so much software everywhere, most of which we never see. Every small business has its little web site. Who knows how much software there is powering this or that device or process at random factories, laboratories, and offices.
The Debian popularity contest looks like it has a big long tail of relatively unpopular packages [1]. It looks like the app store has 2 million apps, only 2857 (.14%) have more than a million dollars of annual revenue. These are of course incomplete and flawed and do not really directly address the question. I don't really know how to research this in a thorough way.
But the vast majority of software is one-off stuff. It makes no sense to optimize it for performance instead of features, development time, correctness, ease of use, etc.