I'm not sure how true this is. I was into Java in '95 when it had "lousy IDE support, buggy compilers, few libraries, few bindings to existing libraries, and so on". Java fell far short of its potential, and the ambitions Sun had for it, but still, that wasn't a bad call. People who got into Python early on made a similarly good call.
And also, some esoteric languages, like Haskell, the ML family, Erlang and so on have deep roots, they may not be mainstream but they aren't going away anytime soon. And sometimes something funny happens; anyone who has ML experience is going to feel right at home on F# and suddenly the library problem is solved, the IDE problem is solved, the bindings problem is solved...
No. Esoteric languages are a fairly clear category of languages created to test the boundaries of language design.
Esoteric languages are not created for real-world usage (or even for usage at all), examples of such languages are INTERCAL, Thue, Unlambda, Moo, Befunge or Java2K. They can be turing tarpits (languages with extremely low number of commands), funges (multi-dimensional programs), nondeterministic (from a given global state A and a current instruction f, the next state `f A` of the system can't be predicted)…
There's nothing esoteric about a language such as Erlang created very specifically to solve industrial problems, or even about languages originally created for research (ML or Haskell), especially when they've now leaked into practical, real-world programming outside of academia.
And also, some esoteric languages, like Haskell, the ML family, Erlang and so on have deep roots, they may not be mainstream but they aren't going away anytime soon. And sometimes something funny happens; anyone who has ML experience is going to feel right at home on F# and suddenly the library problem is solved, the IDE problem is solved, the bindings problem is solved...