From conversation with roddiekieley:

Hey, one caveat I'm not sure we talked about that mikhail just asked about and I'm sure other users will to. During scaling the operator simply requeue's incoming events for when scaling is complete and then processes them afterwords. This is probably unlike other operators, like the router for example, where a config change is pushed in and the router bounced right away as far as I know

However this does mean that if scaling never "finishes" for some reason, like a broker pod can't mount a persistent volume because one is not available, intervention might be required where the right size is set and applied then oc scale is also used to set that size. At that point the operator should note that scaling is "finished" and move on.
Might need some more failure proofing in this area in the future.