Imagine a university gets a huge grant to run protein folding calculations, and the grant lets them do more than everyone else including miners: now they own >50% of the mining power of the cryptocurrency without making any investments intended toward mining and can now do 51% attacks at no added cost.
Or imagine someone with less than 50% of the mining power trying to do a double-spend attack: normally this is discouraged by the fact that double-spend attacks are probabilistic, and the attacker has to pay the full cost for doing the attack regardless of whether it's successful. But if the proof-of-works are valuable outside of their purpose within cryptocurrency as proof-of-works, then the attacker can re-coup some of the costs of their attack by selling the results of their proof-of-works. Double-spend attacks will then have higher benefits and may be profitable to attempt now.
Additionally, proof-of-work systems used by a decentralized cryptocurrency can't rely on a centrally-produced dataset. If the Folding@Home group produced workloads for the cryptocurrency miners to solve, then the group could construct fake workloads that they already knew the solutions to, and then secretly insta-mine those workloads themselves to earn mining fees or to do double-spend attacks on the network.
I think it's better for cryptocurrencies to instead focus on developing solutions that just don't need proof-of-work mining to happen at all, like proof-of-stake.