I think getting "value" vs "objects" right is paramount: this in itself is only a semantic difference yet, but that will allow for a lot of compiler optimizations down the line.

E.g. the value 5 can't be changed, neither in Haskell, neither in C. A place that stores that value can, which makes the place an "object".

Mutability fundamentally means identity (see Guy Steele), so I think this is a fundamental distinction.

As for immutable data, sure, on single threaded code they do carry some overhead, but this may also translate to higher performance code in a multi-core setting due to better algorithms (lockless, or finer grade locking).