I was looking at it while less than half awake, in these terms:
Cumulative score from training sessions originating with level 0 greenhorn with maximum level 5000 trainer, will add exponentially diminishing scores to the trainee, as the differences between scores approaches zero.
Eg, the sum added is the X% of the difference, and since the master has capped, he can't improve from the lesson. As such, the first training session will add "retardedly insane" experience to greenhorn eqivalent to 10% of the difference between the scores. (Difference from 0 to 5000 is 5000. 10% of 5000 is 500. Added score is 500.)
On the next itiration, the difference will be 4500 xp, and so the added score will be 450.
This curve resembles an exponential one, because it compounds the bonus value, albeit in the negative growth direction. Experience bonus approaches 0, as difference between scores approaches 0.
Again, I am basically a walking zombie right now, from chronic lack of sleep. (Mandatory OT + insane-yowling-heat-addled-cat)x5days = "please kill me now."
I was trying, and failed miserably, to point out that a function like this won't misbehave as long as the differences between the scores of the trainer and the trainee are low, because the bonus awarded will be low.
If the implied "normal" situation is for the trainer to not be a heroic avatar of armok himself, but instead just a hardened drill instructor, then the bonus score from the training for a greenhorn will be pleasantly nice, but not excessive, and will fall off to 0 as the greenhorn reaches the same level as the instructor.