A single 910 kW generator is big enough to meet the entire load. When I force HOMER to use an 1820 kW generator with 30% minimum load ratio and a 425 kW generator, during low load periods, HOMER runs the 425 kW generator, but otherwise runs the big one at low load. Modeling the system with two 910 generators and one 425 generator results in only one of the 910 generators running (at 62% capacity factor, about double that of the 1820 generator).

I'll spend a little more time looking at this, but I think it's going to be hard to get HOMER to replicate the generator dispatch. I am getting reasonable total fuel consumption values, which I think is what matters most.

Oh, I see why you don't like the results when you simulate two 910s and one 425: it's because one of the 910s sits idle all the time.  In fact, that's probably exactly what happens in the real system, although the two 910s probably take turns on alternating months or something like that.  Same for the 425s I bet.

I usually model twin tag-team generators as a single generator with double the capital cost, double the replacement cost, and double the lifetime.  So the HOMER schematic would have one 910 and one 425.  The fuel curves, minimum load ratios, and O&M costs would be appropriate for a 910 kW and a 425 kW generator.  But the capital cost, replacement cost, and lifetime would be doubled.

Does that make sense?  I think that mimics reality pretty well.  You buy two generators so the capital cost is doubled.  But you act like you only have one in that they never both run at the same time.  The two generators take turns operating, so the pair lasts twice as long as a single generator would, and when you do need to replace them you need to replace both at the same time, so the replacement cost is doubled.