The exact same thing applies to mining in a pool with different difficulties. Diff 512 or 128 does not matter. On average, you will get paid the same. It does not really matter if you submit 10 x "512-shares" or 40 x "128-shares" (conceptually).
Your "concept" is highly flawed. If you submitted 10x 512-shares and 40x 128-shares, they WOULD be equal. But here's how it works:
Scenario A:
128 A 07:41
128 A 07:42
128 A 07:43
128 A 07:44
128 A 07:45
128 A 07:46
128 A 07:47
128 A 07:48
128 R 07:49
Scenario B:
512 A 07:44
512 R 07:49
What's the difference here? With A you got 8x 128 submitted on time. That gives you a share count of (128*8=1024). With B, you got 1x 512 submitted on time. That gives you a share count of (1*512=512).
Therefore, while your GPU's worked at the SAME speed, you got half the share count.
Now, on things like LTC where the block time is large anyways, this is not such a big problem. But let's look at a fast-block coin:
Scenario A) fast miner:
512 A 08:13 B1
512 A 08:13 B2
512 R 08:14 B3
512 A 08:14 B4
512 A 08:14 B5
512 R 08:14 B6
Now a slow miner:
512 R 08:14 B1
512 R 08:15 B2
512 R 08:15 B3
512 R 08:16 B4
512 R 08:16 B5
512 A 08:17 B6
The fast miner got credit for 4/6 blocks. The second miner got credit for 1/6 blocks. Therefore the slower miner is being screwed out of the vast majority of all earnings. For 5/6 of these blocks, the slow miner got absolutely nothing, despite using power and wear&tear on their GPU.