That's not how that scales though and you should know this.
Plenty of 650-750 MHz SoCs can easily do Gigabit routing, as the MIPS architecture is better than ARM when it comes to routing, for some reason.
On top of that, any decent router SoC has, just like all other Arm based SoCs, dedicated co-processors that handle things like routing, which means the Arm processors are freed up to handle other tasks, like running the OS.
As such, 10 or 25 or even 100 Gbps is feasible on an Arm SoC, I mean how would those big Arm based servers do 100 Gbps otherwise?
It's obviously up to the SoC vendor to test and make sure it really works and to provide software support for the co-processor(s) to make sure it all works as intended. In the past, some companies have charged extra for this, which some router/device makers were unwilling to pay for and the end result was poor performance.
SoC Mhz have just about nothing to do with routing speeds.
I was quoting for SW based routing - without any fast path accelerators...
We get to 10Gbe space, and things get interesting - QCA is up front, and MediaTek is chasing them and close behind.
MIPS was always interesting, as one could run it big-endian, e.g. network order, so it didn't have to do the swap from big to little for packet processing - Broadcom via SiByte was little endian (mipsel) as was Ralink (ramips)...
That being said - you could take a N300 device on MIPS/ath9k, and basically be perfectly balanced between LAN/WAN/WLAN - the AR9531 was that chip, and for dual-band devices, the most recent would do a single stream 11ac 5Ghz channel at 430, which once the wifi overhead was removed, kept the whole device solid at 100 Mbe... as that radio was running on the SoC's PCIe bus
Someone running a pfSense box with a 2.4GHz intel processor can be fine with a 1Gbe connection on the WAN side, and this would be full-duplex on the slow path with all routing, firewall, and NAT running on the CPU.
These days - it's interesting that a lot of the network is offloaded to the NIC's for 10/40/100/200 GBe - nVidia with Mellonox, AMD with Pensando, and intel with their own stuff (which they recently sold) - and there, the benchmarks were a bit different - it was more about packets per second vs. overall throughput - and the PPS benchmarks had a solid figure - the iMix..
More info on iMix here... benchmarks there need to change, as things like Speedtest.net, etc are hard to apply with 10G - they were not designed for this, and as such, will give bad info...
en.wikipedia.org
@TheLostSwede - I used to work in the Telecom Carrier Core Network, so yes, I'm aware of the challenge on 10Gbe on the last mile...
Once we get to 10Gbe connections - things do change, much like Tesla and comparing them to Gas cars on the dyno - the dyno as a benchmark never consider that for an Electric Motor, max torque is at zero RPM, so it runs off the charts and doesn't give a good perspective on what really is...
