It is not known. The rules state:
"If players play under 48 minutes, they will still improve, but less quickly than they would if they got more playing time."
Let's say that you are training two positions, say PG/SG. You have 288 minutes at the two positions, so theoretically you can train 6 players with 48 minutes each. But you most likely will have some players go over 48 minutes, which are wasted minutes af far as training.
But if training were proportional in the manner you suggest, you might be able to have 8 players play an average of 36 minutes. With no wasted training, training 9 players at 75% is equivalent to 6 players at 100%.
So there might be checks to keep teams from gaming the training system. On the other hand, if you are training 9 players, you may be having trouble keeping the players in game shape, and may also be losing games because you are trying to get extra minutes in. For example if you used a different starter in each of the 3 games who played 36 minutes, and the same backup who managed 36 minutes from the 3 games, then you could be using your #1 PG in one league game, your #2 in the 2nd league game, your #3 in the scrimmage, and the #4 as a backup. You may have to be paying more in salaries to get a #2 PG who can start have your league games, and a #4 who isn't totally useless as a sub.
It doesn't make sense that 47 minutes would be drastically less than 100%, so I've assumed that the training efficiiency function is continuous. But that could be satisified with something like: (PT/48) squared. 47 minutes would be 98% efficient, but 24 minutes would only be 25%.