Scenario 1 and 2, standard ringsize
Carol creates a transaction to Dave. It enters the mempool at blockheight 27, referencing the freshly unlocked transaction of the exchange to Alice. Both transactions are included in block 28.
Who can distinguish whether or not Carol is in fact Alice?
Scenario 1 and 2, non-uniform ringsize.
Carol creates a transaction to Dave. She uses ringsize 73. Her transaction enters the mempool at blockheight 27, referencing the freshly unlocked transaction of the exchange to Alice. It also references an output that was part of a transaction with ringsize 73. Both transactions are included in block 28.
For an outside observer, circumstantial evidence suggests that the two transactions using ringsize 73 are indeed from the same entity. Thus, the other transaction referencing Alices input is far more likely to be in fact created by Alice.
This is, if I understand it correctly, the biggest issue @SamsungGalaxyPlayer has with non-uniform ringsizes.
For the record, I’m not yet fully convinced we should make the move to a fixed ringsize in the way we’re currently doing it. If you want to count, count me as withholding my vote lacking necessary evidence in either direction.