On the Numerical Efficiency of C++ in Scientific Computing