Your projection spreadsheet had Brady averaging about 340 yards per game over the next 5. All December games in cold weather sites. All against inferior opponents with the Pats 2 games clear of the Jets/Bills in their division. I just think that is amazingly over-the-top optimistic.
Look back at the December games for Brady. While he does have the occasional 300 yd game, he is more likely to have a sub-200 yard game (for all the reasons I mentioned earlier). Not saying that will happen this year (the Indy game at least should be an exception based on the weather forecast) but the past results do seem fairly consistent year-to-year.
I disagree with you on this, but the point is likely moot. Brees is likely going to run away with the record this year. While Belichick is a student of history, I'm thinking he isn't overly concerned about who is in 2nd place for a particular record...and I highly doubt he puts Brady in harm's way just to end up behind Brees and ahead of Marino.
Leaving your third point aside, you have instinctively put your finger on the problem with using these statistics.
The average of Brady's performances to date is 329.7 yards, but the sample is so small as to be meaningless for forecasting.
To wit, the Standard Deviation of those data points is 93.4, meaning that just going out two Standard Deviations would give a range on possible future performances between 143.0 yards and 516.5 yards per game or, to use your language, would reflect the possibility of "a sub 200 yard game."
Furthermore, if you look at the variation between Brady's individual game performances and the opponents' average defensive performances, there is no meaningful correlation to be found between these two (very small) sets of data (the RSQ is .16, or virtually no correlation). So, using those data for the purposes of projection is, well, meaningless.
So, really, all one can do is look at the circumstances of each game, as others have pointed out, including the playing conditions, the stakes, etc.
If we're still obsessed with the idea of sticking to the "average," but adjust the average for its inherent volatility across just two Standard Deviations, then we would say something like, "Brady will throw for somewhere between 4,342 and 6,209 yards this season with a midpoint projection of 5,275 yards (as another poster pointed out)."
But, that's a meaningless statement for all practical purposes, so we're really wasting our team using these data to make projections that they are inadequate to support.
Best scenario: watch and enjoy the rest of the season and let's not try to use a small group of data to show more than they can show.
(If someone is truly obsessed with this, they could run a Monte Carlo simulation using the available data and then wait until January 2nd to see where the real outcome falls on the curve, but...)