And I fully agree, but we have a lot of "technical debt" in our society. We have to keep supporting our existing "software", and fix all critical bugs, but we should also dedicate some time to refactor the society, even if it requires building version 2.0 from scratch, adding only these libraries that make sense.
As far as optimizing power usage in data centers - armies of engineers are already working on it, and there is really not much more we can gain there, but changes in software, to be more efficient, simpler and less bloated is something that would have a large impact. This however, as in the case with the light bulb, requires coordination of many different people/companies/technologies. This will simply be impossible not because it's hard, but because our society spends too much effort on making stuff look great, and making sure that we can be tracked by advertisers so they can sell us stuff we don't know we need or want.
The problem is, this work adds up superlinearly. The thing about humans is that coordination is hard.
In order to change enough lightbulbs to have an impact on global energy use, you'd have to convince many millions of people to act against their personal, short-term, economic best interest (if it was already in their perceived interest, they'd change the bulbs by themselves). This is a kind of problem you can't really take on directly. An indirect approach would be to e.g. significantly reduce the cost and increase distribution of the better bulbs, leveraging technology and market economy to your advantage. This way the bulbs will start making economic sense, and people will switch by themselves. It's a difficult approach, but much less difficult than persuading people directly.
But then again, you may discover that optimizing power usage in data centres can save as much power, but requires less work and less convincing than the lightbulbs scenario. It would be better to start from that.
This is the kind of solutions we need to prioritize - low effort, high impact.