I think your perception is valid that less attrition would yield better results. The question is really just a matter of scale (attrition likely matters, but how much?).
If without accounting for any attrition, coach effect, home field advantage, or any other factor than talent, one can approach a prediction rate of 70% generally and 90% in title games, then how important can any of those other factors be?
My findings tend to indicate that coaching is the next biggest factor. When you look at any team that performs well above or below talent expectations you begin to find that certain coaches are isolated. That was one of the most shocking things to me about these findings (generally most teams perform to their talent, and the few that don't tend to be touched by a small pool of coaches).
Finally, although I have not done a comprehensive study on attrition, my sample indicates that attrition is actually a very similar range across most major teams. Teams with good talent tend to have transfers (can't get playing time) or players leaving early for NFL, or academic casualties, or discipline casualties at rates within a range that negates attrition as a major factor in my eyes. Yes, Tennessee might be closer to the high attrition range of "normal", but I am not convinced it is freakishly abnormal.
I think that most UT fans believe attrition is much higher because they pay very close attention. Couple that observation with how we fans see historically solid recruiting at UT yields substandard results. It appears logical that attrition would be the culprit, right? Well it might have an impact but looming larger are the fact that Fulmer had a gradual but very real decline in his effect on talent, then Kiff (fired at USC for utter game day negligence) and Dooley (fired by UT for the same). The end result wasn't that UT bled talent at an unusual rate it was that the past decade has had piss poor coaching and more importantly declining recruiting in relation to our major competition. At least that is my take.