In that case, you may as well remove all references to leaves and double SHA-256 from your BIP since your design has no method for distinguishing between internal nodes and leaves.
I think that if this design stands, it will play a role in some future CVEs. The BIP itself is too abstract about its data contents to specifically say that it has a vulnerability; however, I believe it is inviting vulnerabilities.
For example, I might agree with a counterparty to a design of some sort of smart contract in the form of a MAST. My counterparty has shown me all the "leaves" of our MAST and I can verify its Merkle root computation.
After being deployed, I found out that one of the leaves wasn't really a leaf but is instead a specially crafted "script" with a fake pubkey chosen by my couterparty so that this leaf can also be interpreted as a fake internal node (i.e. an internal node with a right branch of 0x8000...100).
Because the Fast Merkle Tree design doesn't distinguish between leaves and internal nodes my counter party gets away with building an Inclusion Proof through this "leaf" to reveal the evil code that they had designed into the MAST at a deeper level.
Turns out my counterparty was grinding their evil code to produce an internal node that can also be parsed as an innocent script. They used their "pubkey" to absorb excess random data from their grinding that they cannot eliminate.
(The counterparty doesn't actually know the discrete log of this "pubkey", they just claimed it was their pubkey and I believed them).
Having ambiguity about whether a node is a leaf or an internal node is a security risk. Furthermore, changing the design so that internal node and leaves are distinguishable still allows chained invocations.
Arbitrary data can be stored in Fast Merkle Tree leaves, including the Merkle root of another Fast Merkle Tree.
Applications that are limited to proof with paths no longer than 32 branches can still circumvent this limit by staging these Fast Merkle Trees in explicit layers (as opposed to the implicit layers with the current design).
By storing a inner Fast Merkle Tree root inside the (explicit) leaf of an outer Fast Merkle Tree, the application can verify a Inclusion Proof of the inner Fast Merkle Tree Root in the outer Fast Merkle Tree Root, and then verify a second Inclusion Proof of the desired data in the inner Faster Merkle Tree Root. The application will need to tag their data to distinguish between inner Fast Merkle Tree Roots and other application data, but that is just part of the general expectation that applications not store ambiguous data inside the leaves of Fast Merkle Trees.