Matlab compiler creating an executable may have some effects on speed (either faster or slower), but as far as my experience goes timings are comparable between a compiled app and code run in Matlab so you are correct on that, to the best of my knowledge.
To estimate speed-up of codegen in a generic manner is impossible though. It depends on many factors, most obviously the code that you are generating the C++ code for. Some code in Matlab is more optimised than others, especially builtin functions. Some of these have gone through many iterations of optimisation.
If you write your own Matlab code with a bunch of nested for loops this will likely be very slow in Matlab, but C++ works fine with for loops so using codegen to turn it into C++ code will likely give a huge speedup there. But it is all qualitative. For all I know some code may even be slower if you use codegen than run the Matlab version if the latter is highly optimised.
You would have to test it for yourself on the code you actually want to use it for.
