The threshold I've read on airliners.net is at about 30 frames. Below that costs of a sub-fleet are too high and you should aim for commonality. Above 30 frames you should pick the best plane for your routes, regardless of commonality.
So for example running two fleets of 30 A350-900/1000, and 30 787-9/10 would be perfectly rational. But having 50 A350-900/1000 and introducing a sub-fleet of 10 787-10 is unlikely to be a wise choice, even for routes where the 787-10 would beat the A350-900.
This isn’t true for the reasons you are implying. Most of these smaller carriers fly to regional airports with smaller runways and smaller demand which the 737 specifically is ideal for (since that’s what it was designed). It’s also one of the most common planes out there, so when smaller carriers are looking to buy, that’s what’s available. It’s not that they are looking to be a single vendor airline. It’s just that the one vendor made the plane they need.
But the specific example I had in mind is Air Tahiti Nui, and its grand total of four 787-9. It’s simply an airline specialized in bringing tourists to Tahiti from far, far away. Hence high density 787-9.
But even a very small airline like Air Senegal has a pretty diversified fleet, of 9 aircrafts…
Edit: in regards to maintenance a lot of airlines are outsourcing maintenance to bigger providers so that’s less of a deal than you’d think as well. Fuel really is one of the largest factors in this and an extra 5 years of expensive gas while waiting for a new plane may be too much for budgets to bear.
Re: pilot training. I could imagine that going from a 737 to a 787 would be substantially easier than from 737 to A320, due to standardisation in interfaces, processes, documentation, etc, within one manufacturer. Is that the case? 5-6 weeks does still seem like a lot of downtime for a commercial pilot, and rolling everyone through that sounds like it would be prohibitively difficult for many airlines. Plus my understanding is that it's sort of a one way street, pilots don't typically (or can't feasibly?) stay rated for two aircraft types for long periods, so it would still reduce flexibility if an airline split its fleet right?
In particular, I've heard that maintenance and pilot training are big issues. If you've got half your pilots on the 737 and half on the A320, you're much less resilient to scheduling issues than having all of them on one or the other. Similarly, I believe maintenance facilities tend to be one or the other, so if you've got half your facilities for each, at the margins you're going to have more scheduling issues for maintenance.
There are economies of scale with anything to do with scheduling flexibility, and dividing like this essentially halves those economies of scale.