The kinds of algorithms you can express very cleanly in a functional language, like Haskell, for example, are a lot more awkward when you have to worry about managing memory at each intermediate step. Look at all the monkeying around you have to do with __block declarations etc. For typical Obj-C code, where you're just gluing together bits of the Cocoa API, I agree that Obj-C's memory management isn't too bad. When you're trying to work at a higher level of abstraction, it's a distraction.
Writing complex algorithms in Obj-C is a pain because of it's incredibly clumsy syntax for collections and the manual coercion of elemental types back and forth to objects. Look at an implementation of a neural net or something as simple as k-means clustering in Obj-C, then C++, then Scala or Haskell to see how poor Obj-C is for this kind of thing.
For example, imagine how much worse this would be in Obj-C:
https://github.com/cageface/brains/blob/scala/src/main/scala...
Admittedly, Scala is particularly nice for this kind of thing, and Java is no better than Obj-C, but even C++ gives you much better tools for abstracting this kind of code.