I agree the recommendation/requirement of no greater than 56A OCP on AC1 doesn't seem right.
I've been considering contacting SMA support to confirm they really meant that, and for their recommended type of breakers.
SI 5048US manual called for 70A max OCP on AC2 input, didn't as for anything on AC1 output. I assumed it could take in 56A from grid/generator and add 40A continuous (5000W from inverter) for 96A continuous output.
SI 6048US manual calls for 70A max on AC2 and 56A max on AC1
Thermal/magnetic breakers are supposed to be run no higher than 80% continuous, to prevent nuisance trips. So 70A OCP is perfect for 56A circuit.
With 56A OCP on AC1, 80% limit would be 44A max, which is less than what SI is rated to deliver.
Magnetic-hydraulic breakers reportedly can be used at 100%, which is why I wanted to see if SMA does recommend them for AC1. Magnetic-hydraulic trip in about a second after moderate overload, rather than sometimes 20 minutes for moderate overload of thermal breakers.
I originally intended to use 70A QO270 breakers on both AC1 and AC2. Because I have 4x SI 6048US and an existing 100A fused knife switch, I stuffed a DIN rail and 2x 63A Schneider Multi-9 2-pole breakers inside it for AC2. Those are also thermal-magnetic so should be good for 51A.
I didn't RTFM (6048US), having already RTFM (5048US) cover to cover, wasn't aware of 56A OCP on AC1 instructions. I put 2x 70A QO270 in protected loads panel and back-fed from AC1. In operation, power handled by inverters reported about 3:1 ratio between parallel inverters. Problem was differing resistance from one QO270 to another.
I disconnected the breakers on AC1, inside wired inverters directly to main lugs of protected loads panel without OCP. That worked fine, with balanced current. I then installed another 2x 63A Schneider Multi-9 2-pole breakers between SI AC1 and the main lugs. That also worked.
18 months later, those breakers tripped. Had been carrying about 25A each, one pair tripped so other pair carried 50A. Within 15 minutes (if not seconds) it also tripped. Warm day.
Those breakers were a used batch. Testing under 60A load, some variation in voltage drop, 80 mV to 120 mV. All held (in a cool garage) except the one which tripped first in use. I swapped in one of the better units and it has been running a couple weeks.
I intend to look for magnetic-hydraulic, maybe 60A.
The manual explicitly states "The maximum input current allowed on the Sunny Island is 56 A. Higher input currents must not be connected to the Sunny Island." (6.3.1) This has me thinking that only AC inputs must be limited to 56A. That is, all PV or other AC sources must not exceed 56A. This would make much more sense in my mind.
With grid backfeed, limit (in US 120/240V market) is 6.7 kW of PV per Sunny Island, because that is 56A x 120V.
Off-grid or otherwise not backfeeding, limit is 12 kW per 6kW Sunny Island. That would be more current, but only to loads; no reason for it to flow back into SI unless there is a fault.
It sure sounds to me that I can forget about using anything more than 20A circuits. What about 30A -40A dual pole circuits for electric ovens, dryers, etc? Are you able to run these in a split-phase Sunny Island system?
That's just a limit on the surge current it can deliver.
SI-6048 "Maximum current (peak value) for 60 ms IAC, max 180 A"
SI-5048 "Maximum current (peak value) for 60 ms IAC, max 120 A"
Magnetic breakers fast-trip at about 5x their rating. 20A would take > 100A, and 30A would take > 150A
If not fast-tripped, lower current like 45A could be carried some 20 minutes before 30A breaker heats up enough to trip. I've used 30A of electric heater to check trip time of 20A breakers, ran about 20 minutes. (will vary with ambient temperature.)
So I think SI-6048 should be able to fast-trip a 30A breaker. I think "trip 20A, not 30A" is left over from SI-5048. If not, inverter would shut off after 60 ms, four cycles of AC line. So you can have larger loads, just know that instead of a fault tripping breaker to disconnect, your power will go out.
If you added a balancing transformer, I think two SI wired for 120/240V would combine their output power, and their peak current, for use on one phase (how well that works depends on how much voltage sag they allow.)
By the way, if slave shuts down, master continues to operate. That means you get 120V delivered to one phase of 120/240V panel. This can be a problem if you have 240V loads, because your 240V loads will be connected in series with 120V loads on the dead phase, and fed 120V to that series connection. A brownout. There is an option (for 3-phase systems) to shut all down in case of lost phase. I need to see if that is available for split-phase configuration. Balancing transformer may also fix the issue.
Still waiting for the answer. Ideally, the manual would simply include some wire-diagrams for common systems, with recommended components, so any electrically-competent person can set it up. 
