Using channels other than 1, 6, 11 (or 13, where legal) on the 2.4GHz band will cause interference on at least two of the 'control' channels (the ones listed) because the 2.4GHz channels other than the ones listed all overlap. That is bad for all neighboring routers, not just the one we're concerned with.
Using channel 1 + 3 (or 11 + 9) for 40MHz channel width would more than likely give all your neighbors a bad WiFi experience if they left their routers on channel 6, and more than likely you too with worse performance than using only 20MHz channel width on any of the main channels available. This is because when neighboring routers are not on the same channel, they can't sync (time wise) to each other and the 'rogue' router is simply seen as interference. However, when all are on the same channel, they can sync to each other and share the time slices as needed so that the throughput of the networks as a whole is as high as it can be.
5GHz channels don't overlap at all and that is why it is easier for the router and us to see a better 'choice' being made in Auto mode.
I still recommend that you take off the Auto channel selection function and take the time to choose a channel manually instead for both bands and all radios. That way you will be in control of choosing the highest raw transfer rates, the best range and the lowest latency channel too for each band and your specific environment and client devices in use. I haven't found any 'Auto' settings that work better, ever.
I test for max raw throughput from a wired client to the same wireless client (preferably a laptop, plugged into an AC power source and with it's settings set to 'performance'). Ensure you test a few locations and orientations (laptop to router) that are repeatable and imitate your real world usage.
I test for ISP speeds with a good speedtest site like
http://www.dslreports.com/speedtest
I also test for responsiveness/latency by loading several large websites after clearing the browser cache and rebooting the laptop at each location too.
It is surprising that the best channel for raw throughput may give the worse latency.
As usual, pick the channel that gives you the best balanced performance for your network usage requirements.
Note that I would avoid using any utility such as inssider when doing the above testing. Not only will it skew the actual results if you do the testing with the utility running, but it will greatly increase the time of the above testing too for no measurable performance increase at all. (Save for having a few more stats available that will not mirror any real world expectations in some/most cases).