Can be a combination of both.
If it's related to updaterate (the 2nd value in net_graph) you should check if the server fps (which is revealed in net_graph 5 as "sv") is dipping under the updaterate.
The third value which you pointed out earlier is cmdrate.
This has a maximum of your fps as you simply can't send out more commands/second than your frames per second.
Which is basically the same as the updaterate dipping if the server couldn't push out more frames than the set tickrate.
Choke and packet loss do come in to play here, there's a few reliable Sourcemod stocks you can use.
I haven't tested them out in terms of performance if you were to query clients constantly or every few seconds, which I also don't see the purpose of.
GetClientAvgChoke
GetClientAvgLoss
Also just going to point out that the servers you're using as examples either use 100 tick or 128.
That's CPU intensive as it is, but combined with the ability of spectating (requiring even more CPU power) and a low nb_update_frequency setting, most CPUs will crumble under the load.
Majority of them also use "generic CPUs'' and are hosted on VPSes/VDSes, which is obviously not optimal.
If you want to guarantee full stability on the Server's side on these extreme settings, you'd have to look at a dedicated machine with one of the latest CPUs (4.2GHz+ to be safe) and be willing to dedicate an entire core to a single server.
As you can imagine, getting a setup like that isn't the cheapest (if you were to factor in networking and possible DDoS Protection).