Based on my benchmarks I suspect that copy elision is disabled. I wonder if the tests are biased against use of STL containers which really need compiler optimization (eg inlining) to shine.
Not only this, but the leetcode results I see indicate that identical code in C++ and C is slower in C++. That means no STL or anything. It looks clear that the options for C++ are screwed up in some way. It doesn't make sense to me to give any weight to the performance of code that's built without being able to set the optimization and codegen flags.