When you compress something, the % of size reduction depends on both the compressed files and the compression algorythm. As an example, an already compressed texture image when compressed (let's say a .JPG with its compression set to max for this example) with that would reduced in size way less than an uncompressed image (let's say a .BMP). And also changes how much something it's compressed between let's say a video or an image versus text files.
A game puts a good amount of different data to the memory, that changes from game to game. So it's pretty difficult to get a "max" or even "average" number of how much this is going to compress the game data. I assume that for these numbers they tested it with different cases similar to the ones they expect to run and did choose for the compressed amount more or less average of common cases scenarios. I assume that a best case scenario for some game may be just a bit higher than that.
A case like the one Cerny said about "up to 22GB/s if particularly well compressed" would be just the result from a test that they may have done by looking at the limits with a file type that is best compressed with that compression algorythm and that they rearranged it to be particularly well compressed. Just more a theorical case than a real world scenario to be seen in a game.
Cerny mentions up to 22GB/s from 5.5GB/s raw data in his best case scenario, which is 4:1. MS claims 8:1 in his best case scenario, but their source raw speed is twice as slow, so the results in both cases would be pretty much the same. And these are best case theorical scenarios. The numbers they show in their specs would be more common case real world scenarios, and there we can see a small lead in compression in MS side, but since raw is speed is also taken in consideration for the final results, PS5 ends being around twice as fast as Series X on average for real world, practical cases.
