parseval.py shows simply how to do the calculation for Parseval's theorem with NumPy's FFT.
The main point is that you have to normalize by the number of samples (depending on your FFT implementation, probably).
Then I made
functions.py to calculate RMS faster in the frequency domain and
example.txt shows the time savings.
Note that there is no benefit to doing the calculation in the frequency domain if your signal is in the time domain. It's only beneficial when you've already converted to frequency domain and done some frequency-domain processing.