It’s because they don’t have the same semantics, you are adding overhead with the Rust version in multiple ways.

I wonder how much difference there would be if I used shared_ptr in C++ instead of the bare one.

EDIT:

As far as I can tell less than 190 lines of asm but I don't know shared_ptr enough to produce exactly the same code.

Yeah, that gets closer for sure. refcell is still going to add a bit of state to the value inside of the allocation, but at least the shared_ptr and Rc/Arc bits are the same semantic, so it'll be closer!

Look at the asm code of rust. You would see that most of the code there is the libraries (and yeah, also because of the semantics, which are powerful).

... the library that implements those different semantics. If you wrote your own Rc and RefCell, that code would still exist, in your crate. Where it lives isn't the root cause.