If the data is application-wide and not user-specific--that's what the application context is *for*. The only real danger is using up memory, IMO. And possibly the need to make sure the app-scoped data is in sync with the backing DB values, but there are many ways around that.
What happens when there is a cluster of 250 web servers hosting a "web application?" Which one of the servers will have the "shared data" Collection object? Will all of the servers have a copy of the data (object)?
I imagine it depends on how it's set up, but I'd be surprised if each machine didn't have its own application context, otherwise there would have to be a mechanism for getting the app scope from a given machine/set of machine, which defeats (one, anyway) the purpose of clustering.