I’m not sure that there’s really a way you can, since AFAIK the Semtech concentrator chips autonomously decide what to assign their demodulators to - the packet fowarder only collects the results, it is not actually involved. Having a second copy of the same thing would only seem like it would help on the random chance that when overloaded they made different rather than matching choices.
What might work would be to assign each concentrator to only four channels, and assign each channel twice, and use the other concentrator for the other four. But I don’t know that would actually help.
Switching to an SX1302 which theoretically has more processing paths might be an option; that’s not officially supported at this point but should be possible as a DIY effort; would probably make sense to test it in a pi first and see if you get any better performance.
Something that’s not clear to me is how two concentrators are actually handled at a system level - two packet forwarders? And if an uplink received on one triggers a downlink on the other, how can their hardware timestamp counters be synchronized. Does that only work in GPS PPS locked mode?
That I’m not sure I agree with; I think you’re more likely to get all of the packets with some distance between the gateways than with them co-located, so that you actually make a meaningful change in which node signals are strong, recognized, blanking others, etc. And you may be able to still receive strong packets with one while the other is transmitting (weak packets probably not; and you can’t even use a diplexor since in your bandplan the downlink channels are the same as the uplink ones)