Is this something I have said, or something you are assuming?
If a low hashrate loyal miner goes through a loyalty-supporting pool, and mines 0.054 coins during a particularly difficult time, the pool would indicate to him he has earned 0.054 coins plus 0.054 loyalty credits. A high hashrate loyal miner solo mines, and finds 10 blocks during a difficult time, and each time he found a block, he sees in his QT wallet, that he has gained 50 coins, plus 50 loyalty credits, for a total of 500 coins and 500 loyalty credits.
Now, the difficulty level goes way down, and it is easy to mine blocks. The low hashrate loyal miner continues mining through the pool, and he pretty quickly earns 0.054 coins and in the process extinguishes the 0.054 loyalty credits. A non-loyal slow miner who did not earn loyalty credits contributes the same amount of hash to the same pool (jumping in for the easy difficulty), but not having earned loyalty credits during the previous high difficulty era, receives only .027 coins, per rules of the coin.
In the meantime, the loyal solo miner, quickly finds 10 blocks during this easy difficulty era, and each time he finds the block, he extinguishes 50 loyalty credits and receives 50 coins. At the end of the easy mining era, he has acquired 500 coins, and used up all the loyalty credits. A non-loyal solo miner decides to jump in to try to take advantage of the easy difficulty level with equal hash power, also solves 10 blocks, but because he did not have any loyalty credits to extinguish as he solved each block, he only got 25 coins per block solved, for a total of 250 coins. Maybe he decides he'd better stick around and mine this coin during difficult times, so next time the easier difficulty comes around, he can earn 50 coins per block. And this is precisely the behavior we want to encourage.
In what way do these scenarios fail to differentiate between slow loyal miners and fast hop miners, or reward the fast hop miners for their behavior?
I like this idea, but it seems difficult to implement compared to a weighted-average difficulty that retargets at a short interval (e.g. 32 blocks) but uses a longer interval (e.g. 2016 blocks) as the basis for calculating the next difficulty so that hash has to increase long-term to increase difficulty significantly.