[Zambyte]

> This however sounds like a million computers computing the exact same thing. Why would one want to do that?

It is what it sounds like, and it isn't ideal. In order for the Ethereum network to reach an agreement on the result of a computation, multiple nodes need to do the computation to check each other. How do you decide how many nodes need to do the computation for it to be accepted by the network? That's not an easy question to answer. The simplest and safest way is to require 100% of nodes to do the computation. Safely getting that % down from 100 to increase the throughput of the network is a problem actively being worked on.

As for a generic use case, think of Ethereum as something like AWS Lambda that happens to have native support for transacting assets.