Looking at the fact that you can do encoding across a lot of different regions seems to me to indicate that there isn't any special capabilities required for Amazon to implement this other than just a custom AMI.
If I had to guess, this is all part of a strategy to figure out ways to utilize their idle fleet. Spot instances are often used for these sorts of batch offline compute-intensive jobs, but if you're a third party you are hostage to the market dynamics (changing prices of spot instances). Since Amazon owns the market, they can just take capacity out of the fleet of available as spot instances instantaneously for free and be immune to pricing changes in the spot-instance market - something no competitor could ever do.
So if you were to ask - what is a common batch-style workload that requires a lot of computing power that a lot of people need who would otherwise not be using our spot instances and that can consume our idle capacity? I can't think of a better market than video encoding.
This is certainly scary for companies like Zencoder, and feels somewhat anticompetitive (IANAL so speaking practically rather than legally), but it makes otherwise good sense as a business strategy.