A pretty basic question, but one hard to define for a Google.

I have two wireless network sub-nets each connecting to its own interface at a site which then connects wirelessly over a 15km backhaul to our gateway.

As it's a pretty inaccessible mountain-top site I'm considering adding a second interface and antenna to connect to the gateway, as a backup in case a card blows and I can't get up to the site immediately.

What I can't work out is whether I would get better throughput with a single point-to-point link (maybe running nv2) with the other link disabled unless needed, or routing each sub-net through its own link connected to the single gateway with a routing rule that will automatically switch everything onto one link if the other fails. Or even if there's some kind of pseudo-MIMO trick you can do with a setup like this.

Is there any relevance in the fact that in my neck of the woods you can pump up to 200w into a point-to-point link but are limited to 4w on a point-to-multipoint? Is it really a point-to-multipoint when you're pumping a 3' beam at just two antennas only a couple of metres apart?

I would say it is still P2P just as MIMO would still be P2P.

You didn't mention frequencies/channels available which has a bearing on possible options.

OK, the links will be 5GHz and there's plenty of room locally, so frequencies won't be a problem.

Here's a schematic. Relays B and C are on the same pole connected by ethernet but each can also talk to A wirelessly.

Is it better to have each network 192.168.x.x connected to the gateway 10.0.1.1 with its own link or to route, say, C to B by ethernet and just use one link/P-2-P backhaul from 10.0.1.2 to the gateway, leaving the C<>A interface switched off unless needed?