PL theory, as a "body of work" has produced a wide range of languages ranging from Agda to Scala to Java. Some of these have actually seen use! Everything from classical music to Justin Beiber. TAPL, for example, is certainly not limited to the less popular languages: it talks about the foundation for Java-style types as well. Some of the same people working on ML and Haskell are also behind designs of Java and C#.
The point of a language like Haskell is to be expressive and useful, not to appeal to a large base population. Popularity and industry uptake are not the only measures of success. (This is, coincidentally, one of the things I don't like much about Berkeley's graduate program, or at least the systems lab I spent a bit of time in: they did seem to think industry uptake to be the only metric that mattered.)
Beyond languages, PL theory serves the role of all theory: it's the foundation upon which everything else is built. Things like the JVM memory model are based on the theory themselves and designed with tools stemming directly from that theory. Sure, the average programmer on the street is never going to use a theorem prover, but they will use the JVM where the memory model has been verified with one.
All this reminds of nothing more than the usual arguments that Linux is a complete failure because nobody uses it on the desktop. But I think that argument is not true even if you limit yourself to "Linux on the desktop is a failure": sure, not many people use Linux on the desktop, but the ones who do find it very useful and are exceptionally productive. Same principle applies to PL theory and functional programming languages.
There are many different ways to have an effect in the world, and the most obvious and direct one is not necessarily best. Even if it feels best.