The non-expert answer is: Because crypto math is difficult for randos to analyze, and it's not obviously the case that running an algorithm 4,000 times is 4,000 times harder to break than running it once.

And from a purely pragmatic perspective, why even consider it? You have a bcrypt/scrypt library, writing your own code is harder and almost certainly worse, so there is literally no reason to spend even a second thinking about it.