The same key is used to encrypt what leaves the iOS device and is sent to the iCloud. Apple gets pre-encrypted data. According to the white paper I read on the subject, Apple then takes the pre-encrypted data from the user, splits in four pieces entangles it with four other users according to an algorithm, encrypts it again with another 256 bit AES encryption to which they DO have the key, and then store it on their or leased servers. When the customer wants his data back, the process is reversed, his data is extracted and sent back. If the government requests data with an appropriate search warrant, Apple will provide them with the encrypted files, but will help no further, meeting the letter of the law. In other words, they get a pile of un-decypherable gobble-de-gook, useless without the users' key and ONLY decipherable on the user's device, where the original UUID is available. (Grin). They have to have BOTH pieces PLUS the algorithm that makes the key. Without all three, either piece and even both are useless. Essentially the white paper stated that any decryption had to be done on the device it was created on.
They also stated there's an increasing delay in each attempt at trying to do it. . . slowing down the process if they do attempt passcodes by brute force on the device. . .
That's what I inferred would have to be the case but didn't realize they'd actually implement that (given that you can't decrypt on another Apple device that you own, etc.).