Could we do finer-grained parallelization

[sfluhrer]

Right now, when we sign, we divide the signature process into tasks; we divide each Merkle tree into two tasks (when it makes sense), and have each FORS tree be a single task.
This parcels out the work nicely if we have up to 8 threads; would it make sense to divide things further (e.g. divide each Merkle tree into quarters, and possible divide each FORS tree into two)? This would allow us to use more threads (e.g. 16) efficiently.
Of course, I'm not sure if this "more than 8 threads" is common enough to merit the extra complexity.