My experience is that popularity does matter a lot, but that the value of popularity is logistic rather than e.g. linear. Not enough popularity does make using a language in production unrealistic, but there is a threshold of usability after which increased popularity seems to have rapidly diminishing returns. I would argue that Haskell passed this threshold a couple years ago for most use cases. There is still a lot of work to do on cabal and some of the core libraries, but we're happily using Haskell in production, and for us the benefits very clearly outweigh the occasional difficulties.