It depends.
A compiled C++ file is not translated to Matlab. You can call it from Matlab like an M-function, but it is a compiled dynamic library.
For a simple a*b, when a and b are large vectors, Matlab calls an optimized BLAS library, which is multithreaded. Of course you can do this from C++ also, but if you use a naive loop, you will be slower - maybe up to a factor of the number of cores in your CPU.
How do you define "faster"? If it is the time to solve a problem, you have to consider:
t_total = t_design + t_programming + t_documentation + t_debugging + t_runtime
Do not measure the runtime only. Only for real-time applications and if the runtime is some days or weeks, the other contributions can be neglected.
