cost - you pay in your own CPU cycles/bandwidth.
reliability - obviously, the startup would have to earn the reputation for reliability, but there's nothing inherently stopping it.
etc - AWS is a beast, relatively speaking, and this offers a lot of smaller PCs for a short amount of time. I can't really think of a reason where that would be needed for computing, but as network relays it would be very useful. You could create your own custom Tor and deploy it on demand.
Yes, kinda like folding@home, just generalized and easy for everyone to use. Also, big advantage is the bandwidth usage (which would likely be a bigger selling point than CPU time).
As for the electricity cost.. The tokens would have to be worth more than what you spend on extra power. And there's also the thing of "why don't I just use my own PC for 2 days instead of 10 PCs for 5 hours each?", to which the answer is go for it if you can. But you may have a problem where you just got the data and need it folded or whatever as soon as possible.
Again, I feel the bigger use case here is the network, which very low extra electrical bill. For compute, you can just rent EC2 or some other compute station in the cloud, but having hundreds of network nodes for short amounts of time is really hard to buy currently.
A botnet startup. People sign up for the service, and install an open source program on their computer. The program can:
For every quantum of data transferred / calculated, the user earns a token. These tokens can then be used to buy bandwidth/cycles of other users on the network. You can also buy tokens for real money (including crypto-currency).
Any job that you choose to execute on the other users machines has to be somehow verified safe for those users (maybe the users have to be able to see the source before accepting, maybe the company has to authorize it, etc). The company also offers a package of common tasks you can use, such as DDoS, Tor/VPN relays, seedboxes, cryptocurrency mining and bruteforcing hashes/encryption/etc.
CIA's The Definition of Some Estimative Expressions - what probabilities people assign to words such as "probably" and "unlikely".
CIA actually has several of these articles around, like Biases in Estimating Probabilities. Click around for more.
In hindsight, it seems obvious that they should.
May be worthwhile to ask this on the Polling Thread.