@lxo because the expensive-to-compute result needs to be easy to verify.
If the correctness of the result isn't knowable by the server without doing all the work again, then it fails as a proof-of-work verification.
That's the reason one-way cryptographic functions are used, and not work such as, say, protein folding.