Given a machine with avx512 (implemented) Merkle root calculations will be cut in half. Given shani (WIP) hash computations should be reduced to 20-25% of native (even with existing optimizations). That produces:
tx.hash(false) 1.5%
tx.hash(true) 1.0%
bk.merkle(false) 0.23%
bk.merkle(true) 0.24%
======================
total hashing 3%