"Performance is ok, I can get the full 500/100MB of my fiber connection, both wired and wireless. Can only shape about 200MB with CAKE though."
I don't get how the sqm speed is so low. If the MT7622 is also clocked at 1.3 Ghz and that shapes about 500 Mbps, what on earth is going on here? Especially if we factor in that the chip should be better in IPC. I was expecting around the same performance.
a53 simply is not a well balanced core... it is a decade old in-order design and I would guess will rely heavily on the memory subsystem, if that is not great a53 will reveal its glass-jaw...
(I really do not want to dump on a53 or Arm, after all it not their fault that mediatek puts this into their SoCs, but I wish commodity all-in-one router SoCs would have used a decent out-of-oder core with a not-terrible memory system as new baseline after retirement of MIPS, but it seems that was not to be, I guess in this market every penny counts...)
Both are cortex a53, both are dual-core, both are ~1.3 GHz (1.35 GHz in the case of the mt7622bv/ rt3200), so don't expect much of a performance difference in terms of the CPU power (well, the wr3000 will be clocked 50 MHz slower) and both are relying on very well optimized networking drivers and hardware flow-offloading into their packet engines. As moeller0 already implied, their raw CPU power isn't bad, but not stellar either.
When it comes to sqm/cake, most of the offloading flies out of the window (hardware flow-offloading is not compatible with sqm, software flow-offloading technically is, but loses most of its offloading potential (so it does make sense to keep it disabled)), which means you're relying on the raw CPU power of its cortex a53 core - and that is not much of a step forward from the previous generation of routers' (mvebu/ ipq806x) cortex a9/ a15 SOCs. Here the exact details of your internet connection matter, plain DHCP is lighter on the CPU than PPPoE, which -in combination with sqm- will have quite an impact on your throughput.
still applies, x86_64 or the RPi4 (cortex a72) are a major step up, compared to the low-performance cortex 53 SOC.
As for the question of mt7622bv vs mt7981b, if you compare them under identical circumstances (big question, PPPoE yes/no), I would personally expect them to perform similarly (the 50 MHz advantage for the mt7622bv shouldn't be that significant), but driver maturity might still favour the older mt7622bv at this point.
It would be quite interesting to benchmark mt7622bv, mt7981a, mt7981a and mt7986a under identical circumstances.