I think it is a useful read for people doing any kind of programming at all, in that it might discourage them from attempting optimizations that seem obviously good, and might have _been_ good, if they were running on 20 year-old architectures, but these days just make the code more convoluted to no gain.

I know that this is the kind of effect the paper had on me (I spent a lot of time carefully reading it shortly after it was published), and now I almost never use what knowledge I gained from it to make performance optimizations, but it does frequently discourage me from attempting optimizations that I now know would probably not have the desired effect.