The short answer is I don't know. However, the convenience of simplifying the calculations is, in my opinion, a win. Again, I am myself not convinced. Probably, my premise on the ubiquitousness of normality assumption on the residuals in ARMA models is itself flawed.
