Wikipedia says about a Merkle tree:
Nodes further up in the tree are the hashes of their respective children. Usually, a cryptographic hash function such as SHA-2 or SHA-3 is used for the hashing.
A cryptographic hash function can be defined using a 1-way compression function. Can we use a 1-way compression function for hashing children? It seems more efficient. We don't do length padding. For every node there is 1 less computation of a compression function.
In Merkle hash torrent extension of BitTorrent and in TTH (Russian version describing Tiger Tree Hashing) they use hash functions, not compression functions. I did not find data on AICH of eMule. What about other tree hash functions (I don't know such; maybe, you know)?