Simon East this is something I’ve wondered too. It’s probably so that you can reduce the overhead of having EC2 instances on low load while others are under high load.
If you have a cluster of EC2 instances for each App, Apps under light load have a high overhead.
If you have a large pool of ECS bound EC2 instances, the apps can all share the resources while still being jailed off from each other. You can then also scale the entire pool based on metrics rather than scale based on a particular apps metrics.