1 thing I found lacking: discussion of optimal power point tracking. If ignored (PV panel -> load directly) then depending on the load's characteristics, it can pull the solar panel into very sub-optimal power generation.
That is an advantage of heaving batteries in the system: they can serve as a predictable / constant load, which a charge controller uses to have the solar panel operate at near-maximum output.
But as the article notes: any power that goes through a battery rather than used directly, is relatively expensive. So it makes sense to minimize battery size & throw money at more solar panels instead.
I think there's something to be said for time-shifting predictable large energy consumption events to points at which you're producing more solar energy than can be stored.
Scheduling an electric water heater to run at noon until 2pm for instance, depending on the insulation and size of the tank might be enough to provide all the hot water you might need for 24 hrs. Likewise, for electric car charging.
But there is definitely a minimum battery bank size requirement for maximum efficiency, especially if this is an off-the-grid setup where you can't rely on the grid to act as infinite excess storage.
Without batteries, you're either overproducing (and therefore throwing that power away by backing off the MPPT point), or you're underproducing (and therefore browning out). Therefore, you have to size your solar array for the worst case spike. AC locked rotor might be 30A for instance -- better make sure you're producing at least 7.2kW (~9.5kW of solar+) at any time the AC might flick on.
The higher the peak-to-average of your daily load, the more inefficient your setup will be. In our 9.5kW array example, this means any time you're consuming less than that, you're paying for panels that are effectively offline.
I think it's an interesting problem to solve, with a lot of variables. If the limiting factor is $ vs. roof space vs. ROI vs. 99.9% uptime, you might get different "optimal" answers.