Yes, using historical returns to construct a distribution is used as well, Vanguard does that here https://retirementplans.vanguard.com...estEggCalc.jsf
It just shifts the issues around somewhat unfortunately: 1) Like you point out, who is to say that distribution is an actual reflection of the underlying future distribution? It may be a better guess than gaussian or any other distribution, but you have no way of knowing that and 2) Well, as Vanguard says it themselves:
In particular, the Monte Carlo methodology used here assumes no relationship between asset-class returns from one year to the next. Randomly selected years are considered in sequence. For example, in a given simulation the returns on stocks, bonds, and short-term reserves for 1982, when the nation was deep in recession, could be followed by the returns for 1999, a bull-market year.
So you can have a simulation with five -30% returns in a row for example, something which hasn't happened in reality and is equal to an armageddon scenario. Likewise you can get some spectacular outcomes if crazy bull years are repeatedly selected.
I personally get more mileage from a combination of Firecalc and top level economic expectations.
In the long run growth comes only from economic growth and dividend yield. At the world level that's been surprisingly robust at 3% to 4% in the past 100 years or so.
In short, I'm counting on more short term volatility and sustained long term results. The last part because I'm a optimist, the first part because of technology.