It does depend on your detailed requirements and system load. But overall I would recommend to go the Active/Active approach since its somewhat easy to implement.
The benefit of this approach is a slightly reduced latency then, since under normal conditions the load is shared between the individual units and will therefor result in more "idle" time to pickup new connection/request.
But keep in mind that you don't overbook the individual units, so that in the case of a system failure a single unit will be able to handle the entire load.
If you're running an F5 environment with 3 units, then Active/Active/Standby will be ideal. In this case the third unit can be configured as hot-standby for both Active units. In this case you can book the Active units up to 100% load without getting problems if a failover (single device failure) ocours.
Cheers, Kai