It looks like you are on a cable/docsis network that does not use ds-lite (as seen from the MSS 1460 value in the speedguidenet results). So for all we know your overhead is 18 bytes (and with some bad luck 22 if your ISP uses a VLAN), so reducing that sounds like a decent first step.
I would also add the following:
option linklayer_advanced '1'
option tcMTU '2047'
option tcTSIZE '128'
option tcMPU '64'
option linklayer_adaptation_mechanism 'default'
This will handle the fact that on DOCSIS network we will always send ethernet frames and these are at least 64 byte in size (otherwise cake might underestimate the size of small packets leading to weird sporadic latency spikes)
I would also add:
option iqdisc_opts 'nat dual-dsthost ingress'
option eqdisc_opts 'nat dual-srchost'
That will first share bandwidth fairly between active devices and should help to isolate your gaming PC from other users (within reason, if say 200 users are concurrently trying to download a debian ISO image each will only get ~1Mbps, but for fewer users this typically works well).
I believe that triple-isolate is the default, so you can simply use the better maintained piece_of_cake.qos or layer_cake.qos.
These look a bit weird:
with your settings I would expected the following good-put:
170.00 * ((1500-20-20)/(1544)) = 160.75 Mbps
29.75 * ((1500-20-20)/(1544)) = 28.13 Mbps
But you only got 106.2/22.09
Since you own a r7800 router have a look at:
https://forum.openwrt.org/t/r7800-sqm-settings-keep-causing-bufferbloat/39720/9
and
https://forum.openwrt.org/t/r7800-performance/15780/51
for hints how to configure this for traffic-shaping at high rates. I do not own an r7800 and hence can not really help with the details.
As expected you selected the correct WAN interface.
The deslreports speedtests show:
a) sqm improves your uplink/egress considerable (lower average bufferbloat and less variation)
b) something seems off with the downlink/ingress as you basically only get a little more than half of what you see without sqm, and the average bufferbloat is higher with SQM active (albeit with less variability).
I believe b) is a symptom of your router's CPU being overloaded. But not because the CPU is to weak in general, more likely due to the CPU falling into powersave mode and nnot being able to react quickly enough to the variable load cake processing. The linked threads have some recommendations how to remedy this. As a first step you might want to set the download shaper rate to 0 to disable downstream shaping while you research the linked threads.
Good luck.