That's the whole point of letting the GPU do it though.
But while the GPU is doing that it has less processing power for other tasks. If the edit isn't too tasking you won't notice a difference but if you start stacking the effects and pulling multiple streams of video your performance will degrade faster if you are editing AVCHD footage vs ProRes (for example).
I could say that it is a matter of semantics, but it is more than that. If decompression happens on the fly, then decompression happens. It is the decompressed video that you are editing. Decompressed video is not native. When Apple says that you can edit natively, then I would not call it "lying." I would call it "marketing."
Decompression always happens on the fly so you can view the images. DVDs are decompressed on the fly. A JPEG image is decompressed for viewing. ProRes, DV, DVCPro HD, etc., are all decompressed on the fly the moment you hit the 'play' button. And if you are trying to decompress a very complex codec, like H.264, on an old computer like a G4 the playback will be horrible because the CPU can't decompress the video stream fast enough for real time playback. There is a difference between decompressing on the fly for playback and transcoding from one codec into another.
Let's say someone gets a letter in Morse code and reads it aloud in English. That person is 'decompressing' the letter on the fly but the letter itself is still in Morse code. That is different than if someone received a letter in Morse code, wrote down the decryption in English in another letter, and then read that aloud, right?
Lethal