Do you? Throttling really only asserts that the relevant bottleneck is under your control, but does not necessarily solve the typical over-sized and under-managed queue problem... for that you still should use competent AQM on the throttled link. For egress OpenWrt mostly does the right thing for you, by defaulting to BQL (on ethernet devices supporting it in the driver) and fq_codel as qdisc, but that will not really help for ingress traffic...
Please do not assume malice or incompetence, but always first assume a benign communication issue that can be resolved amicably.
That last words saves that paragraph. All browser tests are compromises to a degree, and not all are equally well suited to each individual internet access link, but that does not make them generally "useless", which you do not claim thanks to the final "for me". Saved off the line ;), smooth!
I don't think I'll get any better........
Should I change my cpu governor? If I change to ondemand, there's a bug and my router turns off/doesn't reboot. I tried sched and that's fine.
Until a proper fix is decided upon and implemented upstream, users should set the minimum frequency to 437500 Hz up from the default 300000 Hz in order to avoid a potential to hang on boot caused by firmware bug.
i don't know exactly what settings you changed, but i don't see a big difference. so i conclude that you have reached a level where whatever you do, it doesn't change anything anymore.
At least I don't have 100+ latency now. and the 'tests' are always the same now.
I don't have packet steering on, IRQ isn't installed, DNS is set to 8.8.4.4 and 900mbps download for SQM and 45mbps upload. fq_codel and simplest_tbf.