RAK has options for a16-channel gateway vs an 8-channel. We’re deploying RAK7289s with 16-channels even though we’re still only using the 8-channel US915 SB2. In theory, someday when we’ve eliminated the 8-channel GWs from our deployment, we could change to 16 channels (but this someday may never come since we mix 8-channel local GWs such as LR9s into our network).
IIRC, proper LoRa Alliance certification for US915 requires motes to support all 64+8 channels - at activation/join time, the LNS sends down the channel map with channels in use (such as SB2). The downside is potentially longer activation time as motes step through the sub-bands until they receive a join accept. In the case of SB2, it’s not terrible because the mote starts at SB1. Our RAK4631-based motes are limited to SB2; our RAK3172/STM32WLxx motes use the ST LoRaWAN stack and do not use hybrid mode.
Besides spreading traffic over more channels, 56- or 64-channel GWs permit higher TXpower in the US - LoRaWAN US915 regional support limits TXpower to +20dBm when fewer than 50 channels are in use (I think this might be a too-literal reading of 47CFR15, but I digress). With 50 or more channels, max TXpower is +30dBm. Even with one of the high-power modules (such as those from eByte) that are capable of +30dBm, the LoRaWAN stack limits TXpower to +20dBm as a result.