I think you're being premature in the view that it "hasn't
taken off" and I'd say 2-3 years from now the landscape could look quite
different. Certainly I can see Scala joining the top tier of language with Java, C#, etc. though probably never overtaking java completely. I know for my company at least we're just hitting that tipping point where Scala is now the first choice for new projects.
Scala, Clojure, etc. are still relatively young languages and we're only just reaching the point where the quality of tooling and learning resources is good enough to make them serious commercial contenders. Tooling in particular is a big part of it - I personally find that I can still write java code much quicker than scala a lot of the time even though it's much more verbose, just because the IDE is so powerful. It's only in the last 12 months that Eclipse scala support is ready for prime time, and even now it's still missing useful things like documentation-on-hover that I take for granted with java.
I guess there's also the underlying problem that OO languages, java in particular, are still the standard being taught in university (yes, you might have one or two modules on FP but it's probably not what most coursework is done in). As the previous thread showed it's very hard coming from that background to switch to an FP way of thinking and I'd say most developers aren't going to want to put in that level of effort in their own time.
I attended a interesting talk recently that the BBC's tech department
put together on FP in online media. They had a panel of media companies
who are using FP languages to go through the reasons why they were
making the switch and the challenges they found. There were a lot of big
names there (ITV, Sky, Guardian, Daily Mail) and they all had pretty
similar reasons - problems of concurrency, problems of scaling, etc. A
couple of them had brought in "FP coaches" to help with the transition. I
can see this approach becoming a lot more common too, though probably
not to the same extent as aglie coaching in the last few years.
One place I can definitely see FP use expanding quickly is in big data processing. We've recently switched our analytics stack over to scalding (twitter's scala map-reduce framework) and it's been a massive success compared to our previous approach using Hive and java UDFs. Map-reduce tasks are very well suited to an FP - there's typically no real concept of "state" and it makes the jobs much easier to test locally before trying them out on a live cluster (very important if you're running clusters with[masked]s of nodes for multiple hours). It's also allowed us to make the analtyics codebase just part of the general code that any developer can work on rather than requiring a dedicated big data expert.