Following this thread intently since I'm in the same situation AND I have some of the same comments and feedback.
About two months ago I was a Single router RT-AC68U user with a single SSID, fixed channels (11/20 and 149/80) and roamassist enabled on 2.4Ghz only, with a value that worked for my house and yard. My reliability was stellar, no disconnects, no offline devices... Nice. I had a single guest WLAN for my 4 IoT devices - some cheap smart bulbs. It also helps to have only 2 distant neighbors and an uncluttered spectrum.
About 2 months ago I 'upgraded' to two RT-AC86Us using AI Mesh AND Smart Connect which throws all that manual stuff out the window. I've started having issues with my IoT devices, showing offline in the vendor's Smart App. When this happens, I can see the WLCEVENTD flooding the logs as it keeps (unsuccessfully) trying to reconnect; I know the log entries are NOT the source of the problem. Restarting the wireless service or disable/re-enable the Guest network will allow them to reconnect temporarily.
I disabled SmartConnect only - leaving everything else to auto (and keeping Same SSID) for the past 2 nights and so far the devices have NOT gone offline. The byproduct of this is my 5GHz capable devices, once they decide to re-associate with the network (probably due to low signal) will connect to the 2.4Ghz network and never go back to 5Ghz. Hence the benefit of SmartConnect.
Now, I just set the channelspec nvram variables as mentioned above (thanks!) with one exception, 2.4Ghz to only 20Mhz - I think that's actually the consensus/recommended value for a number of reasons we should all know. We'll see if SmartConnect ON with Fixed Channels solves my IoT offline issue.
One observation I made is that after I turned OFF SmartConnect two nights ago, that annoying dcd fault started occurring. Prior to that change, I have NEVER had a dcd fault. As the endless discussion about the crashing ended I figured Asus fixed it... but regardless, I dont recall seeing any discussed correlation between SmartConnect and the "Feb 12 08:48:01 kernel: dcd[15536]: unhandled level 3 translation fault (11) at 0x00000000, esr 0x92000007" errors, so I'll just mention it for reference.