Colin pretty well spelled out where I was leading you, but here are the answers to your specific questions:
@fjulianom
@cclemmer Users will not use critical applications or voice over Wi-Fi, so I think a hard roam is not a problem. But how does this influence in the cluster design? In any case, I can configure more than a cluster and enable L3 roaming between them.
Why add the additional complexity with L3 roaming if a single large cluster would handle the scale needed?
At the moment they have AirWave, but if they want to change, Aruba Central can also manage different clusters, right?
That's correct, Airwave and Central can both manage different clusters.
Do you mean is easier to track a user within a big single cluster or within one of some smaller clusters?
What types of problems typically need debugging? Usually roaming, since stationary devices tend to behave fairly static. If a device is also roaming between clusters, now there are different management points to check in order to get a clear view of the client behavior.
My response was intended to stir the thought that a cluster deployed on a single floor may not be restricted to just that floor. If a client device can see APs on neighboring floors, it may well decide to try to roam to one of those APs. So the decision to break a cluster into multiple smaller clusters should consider the benefits vs the added complexity. There not necessarily a right or wrong answer to the question, it comes down to the requirements and what the customer can handle when it comes to daily operations and management.
I've over-engineered solutions for customers and needed to go back to simplify the design to better fit their operations. Trying to save others the same. ;)