Your first paragraph is about laziness-vs-strictness, not functional vs imperative. Most Haskellers would agree that laziness is probably the least important benefit of the language.
It is true that the flip side of GHC's impressive optimizations is the performance of code is harder to reason about just by looking at it. One could argue that this is an simply unavoidable downside of having powerful optimizations -- the simpler the compilation process the easier it is to reason about performance.
Last I checked one can simply import a foreign function as pure, so they wouldn't need unsafePerformIO. Still, I wouldn't consider applying unsafePerformIO to a function that actually is pure a hack.
Regarding (hard) real time, check out http://stackoverflow.com/a/1267814 . Keep in mind that is 2009 so the state of things today could well be a lot better.
It is true that the flip side of GHC's impressive optimizations is the performance of code is harder to reason about just by looking at it. One could argue that this is an simply unavoidable downside of having powerful optimizations -- the simpler the compilation process the easier it is to reason about performance.
Last I checked one can simply import a foreign function as pure, so they wouldn't need unsafePerformIO. Still, I wouldn't consider applying unsafePerformIO to a function that actually is pure a hack.
Regarding (hard) real time, check out http://stackoverflow.com/a/1267814 . Keep in mind that is 2009 so the state of things today could well be a lot better.