So there are some domains where that parallelism easing can really work out well, when dealing with jobs over thousands of cpus, or gpus etc. It can however get almost comical in the case of desktop/single server applications, where you have maybe 4 or 8 cpus available. Because you can write something nice in F# or Scala, and run it on 4 to 8 threads, get a nice speed up. But the same code, written in an obvious manner in C, ends up auto-vectorized and/or auto paralellized to some degree and runs ~8x faster on just one core. Could be more, of much less, depends on many factors, but it is something to consider.
This isn't an inherent problem, but most functional languages need a lot of compiler/JIT work to make the parallel features a net performance win more of the time.