This a topic that has been discussed a lot, I think. There are several threads in the kubernetes users Google group that are worth reading, IMO.
In my experience, it makes sense to have different clusters when you have some hard isolations requirements (there are some limitations today, that are being addressed step by step) or when you want to upgrade kubernetes and not impact all on the cluster or things like that.
For example, we use different clusters for staging/production. One reason is that: don’t impact production on kubernetes upgrades. We have more reasons (they run on different accounts, etc.).
If you would have “weird” requirements for isolation, you might be better with different clusters.
Would you like to share a little more?