Right. I think, in general, PFP (pure-FP, i.e. Haskell and a few more obscure languages) sells itself as a solution to a problem and some people start believing it's the only solution to the problem, rather than one of many (and probably not the best one, at that). Not that Haskell's virtues aren't real -- they are -- but they come at significant cost, and are often available come much cheaper with other approaches. For example, referential-transparency-everywhere is just one of several solutions to the problem of mutable shared state, and, IMO at least, not among the best solutions. Even when it comes to provable correctness, which Haskell touts as a relatively unique feature (at least relative to mainstream languages) people forget -- or more likely, don't know -- that there are fully verifiable imperative languages out there (with their own serious problems -- correctness is never free), that have actually been used by the industry to write some very important, safety-critical software, and have been actually put to the test much more than Haskell.
In fact, one of Haskell's marketing advantages is that, after 20 years, it is still at the stage of selling the dream. Because it has hardly been used where its strengths are supposed to matter most (complex programs employing a lot of concurrency), it gets to sell its "promising initial results" without having to contend with too much disappointment, which invariably arrives once a technology has been well tested. It's the promising NBA rookie who got picked first in the draft but has never played a full quarter, getting people to say, "if only they'd get him off the bench...".