Choosing to do a power spectrum analysis with already epoched 2 sec segments is equivalent to doing in it in a continuous file ?
No, because there is no overlap. If you can, compute this from the continuous files.
What is the main advantage of doing a welch as I did instead of an FFT averaging 2 sec epoch files ?
There is no advantage, this is the same thing.
It would be concerning if the two approaches gave different results.