I was thinking more on the lines of containerization vs virtualization vs bare metal in regards to the efficiency. A compute cluster running a stack of containers all backed by a storage area network can handle a pile more transactions than a bunch of desktops all running their own OS with dedicated drives.

The distributed spare cycles thing works well for projects like the FAH or BOINC systems where all are working on a singular project in small pieces but less so for other things just because of the transfer and transaction overhead. Still, it's a fine way to start putting some control back in the user's hands that can be run with a minimal investment.