>A system where anyone can rent out their CPU time will result in less centralization than we see with current cloud providers. This is an aspirational wish but not reality at the moment. Nobody including BOINC or any other decentralized grid computing endeavor has proven that economically. For your cause & effect of "less centralization" to happen, homeowners cpu would have be cost competitive with cloud services. Selling "idle" time on home cpus isn't cost effective. Most realistic computation workloads also require large datasets (diskspace) in addition to cpu cycles. People don't pay only for the Amazon cpu clock cycles; it's also S3 data storage and network bandwidth.
Likewise, Filecoin's whitepaper for decentralized disk storage is also aspirational and economically unproven at this time.
>A system where anyone can mine productively will result in less centralization than we see with current cryptocurrencies.
Probably not. The "mine productively" in your solution has 2 components: (1) the useful computation and (2) the cryptocurrency computation.
For a thought experiment, let's assume you invent the 95% useful computation algorithm that you targeted. The 5% computation will still be have to be done on hardware where different entities can spend vastly different amounts of money. (Pay more for more powerful computers or pay less for watts of energy because China computer is near hydroelectric dam, etc.)
The 95% of useful computation (say brute force protein folding or DNA cancer analysis) can be an absolute gain for society. However, the 5% computation's "productive value" is _relative_ to all other participants mining that cryptocurrency. Since others can work through that 5% computation much more efficiently (both in pure hardware cycles and/or energy efficiency), the 5% crypto work on the homeowner's laptop loses value in relation to everybody else. Whatever crypto coins the laptop successfully mines, others have mined even more coins rendering the laptop coins to be worth less because of the relative time & energy it took on that modest laptop cpu.
The 95%/5% ratio that's resistant to ASICs does not stop the inevitable trends towards centralization into major entities holding disproportionate power and/or coins in the system. The 5% "non-useful" computation can still be optimized by people spending more money than others. Since spending is not decentralized, the resulting rewards from the 5% computation will also not be decentralized.