I'd say switching between languages within the same paradigm has a pretty low payoff. But if you switch between paradigms -- say, you learn Prolog when all you knew was C -- the payoff is pretty big. This is because you're not merely learning new syntax, libraries or toolchain (boring!), you're learning a completely different way of thinking about problem solving.
Of course, some people love programming languages, and as an intellectual pursuit, studying them is as good as any other. But learning about, say, mechanical sympathy, data structures, concurrency, concepts in distributed programming, compilation or pretty much any CS topic which isn't tied to a specific language would have a greater impact for less cost. It's certainly good to know one language at each "level" -- systems, applications and scripting. Other than that, it's a good idea to spend as little time as possible on new languages.