How would that work anyway? I haven't read the previous 7 pages but if all blocks are compressed with an unpredictable and thus random compression ratio, you would have to set up a whole meta-filesystem in which the AFF would have to reside. I would imagine that some disk accesses would slow down a lot because you would have to read some kind of an index map first in order to know where your block is located on the harddisk. Unless caching is implemented, this would require two random disk accesses for reading a single block. Relocating the heads on a harddisk surely takes quite a bit in comparison to the time saved by reading only a couple of compressed blocks.