Lebzar

My setup consisted of:
Victron Multiplus II 3kVA
Fronius Galvo 2.5 on Multiplus output (with 9 x 250W panels)
Victron MPPT 150/45 (with 6 x 360W panels)
Victron Cerbo GX
3 x FreedomWon Lite 5/4
Carlo Gavazzi ET112 meter on grid input
I built this system up over time, slowly adding more battery, then more PV panels then more battery again and reached the point where I needed more inverting power.  I got an extra Multiplus II 3kVA, studied the parallel installation documentation and watched the training videos over and over.
One of the main considerations when doing this is, that electrical cabling must be as close to the same impedance on each inverter in the parallel system as possible.  This is achieved by using the same type of wire and making sure they're the same length from each inverter.  
After I had installed everything, done the necessary firmware updates and reconfiguration I fired everything up and it all appeared to be working... until the next day.
My grid setpoint is set to about 10 - 20W and for the most part this is fine.  There is occasional brief grid feedback when a large load (eg. geyser) is turned off, but it stabilises pretty quickly.  On day 2 I noticed that when my 2kW geyser (wired on the inverter input side) switches on, the Fronius PV gets cut back to near zero and the Multiplusses started inverting to the level that they were taking all the DC coupled PV and draining the battery. 🤔
I switched the "feed excess AC coupled PV to grid" setting on.  The result was the Fronius ramped back up to full power, the load was carried and the batteries getting some small charge and now the whole system was feeding 250W back to the grid.  🧐  Switching the feedback setting off, resulted in the Fronius cutting back and the batteries being discharged again.
After some head scratching and rechecking of my installation I found that while I had paid meticulous attention to ensuring that the cabling on the AC outputs of each of the Multiplusses was as identical as I could get it, I had not paid enough attention to the AC cabling on the AC input of the Multiplusses.  I pulled them out and measured the one at around 2m and the other at 1.5m.  
I cut the longer one down to the length of the shorter one as accurately as I could (probably within 5mm) and reinstalled.  Voila, problem solved.
So why did this problem manifest in this way?  The documentation all states that if cables are not identical, the load will not be evenly distributed between the inverters.  It says nothing about grid feedback or AC coupled PV being affected. 
The best explanation I can come up with is: the slave inverters input cable was the shorter one, meaning that it was the one feeding in more energy.  The master was the one determining the load and deciding what the energy feed in level should be (per inverter).  The result was too much energy being fed into the grid side of the inverters.  The ESS running on the GX then saw grid feedback in excess of the grid setpoint and decided there should be less power being fed back.  It then chose to cut back the Fronius output instead of the Multiplusses.  The Multiplusses dutifully complied with the power request from the GX by drawing from the battery when the DC coupled PV was not sufficient.
It's a bit of a complex interaction and a quirk of ESS that I hope Victron will address at some point.  If the battery is not full, there should be no reason for ESS to cut back the AC coupled PV output, and even if the battery is full it should cut back the Multiplus output first before cutting back the AC coupled PV.