Why do Intel protected file system makes key per block, but not per file?

Now I m looking on the code of Protected_FS(Intel Protected File system).
IPFS has a merkle hash tree to store (key,mac) of child data node, which is a block(4KB).
It means that evey block should have own derived key which has temporal, spatial overheads.
I think the IPFS can have only one key per a file which will have much less overheads.
Then, Why do IPFS have key per a block? Is there any other reason why do IPFS doesn't store only one key per a file?

Start by reading the Protected_FS implementation and its IPFS Merkle hash-tree handling for 4KB blocks. Trace how keys and MACs are associated with child data nodes, then determine whether the per-block design has documented security or overhead trade-offs compared with one key per file.

Written by the indexing model from the issue text.