I've run through a 20+ season study and I see (at least partially) what's happening. I'll be feeding Markus some detailed data here later today. The basic issue is that the average talent in feeder leagues rises quickly each year for about 8 seasons, then hits an equilibrium--so feeder leagues player creation algorithms are not well-tuned to their levels. After about 6-8 seasons, everything comes to an equilibruim and things seem to work out okay, at least.
One interesting chart that indicates what I'm saying is the average age of major league players when viewed over the seasons. It takes a dip for a bit, then recovers....
Age by Season:
28.1682
27.9994
28.1858
28.3087
28.1618
27.7311
27.789
27.6101
27.6443
27.776
27.5768
27.6394
27.8492
28.1211
28.4033
28.4232
28.5763
28.8665
29.2188
29.1241
29.1075
There's more to it--everything is more complex than it first appears.
But the good news here is that the algorithm is not fundamentally broken--but it does appear to need some fine-tuning. I'm guessing Markus can improve this fairly easily if presented with the right data. I will keep working on this today and post more as is appropriate.