First, I've never seen that huge of change for one pitcher, and don't know what stadium, what league, opposition faced, etc, he was in; plus it was real time, which throws additional variance. Additionally this was many years ago, and we can't verify Randy's numbers in anyway.

For a 2.5 ERA guy, I'd predict a ERA in a 20 team replay to be approx 3.25. Pitchers have a higher variation, up to 50%, due to the following reasons: platoon lineups, type of defense behind the pitcher, pitchers going when they are tired, being left in the game longer than they should, etc.

Additionally you have the lefty effect, which might add another 1/2 run. So I could see Johnson haveing an ERA of 3.75 ± 1.5 runs before park effects were taken into consideration. Thus Randy could have a 5.25 ERA based on stardard variation, and depending on the park he played in, it could balloon higher another 10-20%, putting it close to 6.

I'm also in a DMB league, and have run many season long simulations, mainly becuase I heard that DMB was the most statistically accurate game on the market, even employed by ESPN. I also like the fact I can reproduce a whole season in 10 minutes. You get the same type of fluctuation with that game as you do here.

It is statistically impossible to have a player ERA or batting average be within 10% based on these sample sizes(200 innings or 600 atbats).

With these sample sizes you will always have fluctuation(approx. 20%) that will carry a .300 hitter all the way down to .240 or up to .360 or a 4.0 ERA pitcher between 3.0 ERA to 5.0.