There's a preference for C99 complex when performance is desired from compiled code, but that doesn't affect the performance you get from someone else's library.  If you are forced to compile your own code under a prohibition against C99, it's possible that declaring your complex arrays as plain float or double may give better performance.
You'd have to dig into the MKL documents to see what it says about which functions take advantage of threading for which problem sizes, if that's what you mean by your talk of multi-core mechanism.