Hi again hmani,
I forgot to reply to this part:
you should expect a settling time for the power computations to reflect the correct value. If you can monitor the RMS measurement after the reversal for 37.6 ms, you will know the approximate signal level that the IC comprehends at that point.
The RMS subsystem has an extremely long settling time (440 msecs according to the datasheet). That's fine because the RMS values don't really matter for anything. I would imagine the energy subsystem would settle much much faster than that, otherwise we'd never get accurate energy measurements on very dynamic loads, or are you saying that 440 msecs settling time applies to the energy subsystem as well?