That seems very odd - if it's possible to make those optimisations without any additional type data then why wouldn't GCC do that anyway? The benefit of stricter type rules is that more information is available to the compiler. Using a different compiler doesn't inherently increase the amount of type information.
I believe the claim is more precisely stated as "Many C programs are valid C++ and are faster when compiled as C++" - i.e., even though the text of the program didn't change, the rules for interpreting that text changed, and it's that difference in interpretation that permits better optimizations.
