However, the figure eight follow through and flat figure eight are not the same thing, so it doesn't make much sense to me to make deductions on one based on the other.
Some confusion here as I feel the term "flat figure eight" is misused, or at-least mis-understood. Lets break the variations down:
1)
Figure Eight Follow Though (or retraced

- standard tie in knot, this is the knot we are concerned about loading from within the loop formed
2) For lack of a better name, the
Figure 8 EDK- this is the knot that proved fatal when used in place of a EDK, and is tied just like the EDK but by adding one more twist...
3)
In-Line Figure 8- This is a very strong but hard to untie knot that looks just like the other 8's, but is vastly different. To join two ends you would first make the "8" in one end, then
re-trace the 8 from the
opposite end. The big difference is you don't just grab the rope ends and wrap around EDK style. This knot would actually block any possible rolling, but is not feasible as a tie in as it REQUIRES both ends of the rope to tie... It is this knot that I have also heard called the "Flat Figure 8" and should not be in this discussion as it is impossible to employ it as a tie-in...
So, JBrochu, it is absolutely prudent to make deductions from what Tradmanclimbz reported, the two knots in question where #1 and #2 above. While created differently, they are structurally the exact same when loaded within the loop, but in the case of the rappel error the "loop" doesn't exist, the knot structure however is identical, and not intended for a strong load (even failed under body weight rappelling).
That being said, if it rolls in your described scenario a backup knot would prevent failure... You know it's this weakness loading an 8 in this manner that is why we use butterfly's on a bight or direction 8's in glacier travel... they don't have that weakness...