It doesn't seem that it's the correct metrics, none of them makes much sense.

Something is definitely wrong with the "Queries per second", it doesn't look like something significant. My usual load for a single node is about 300 req/sec. That's just an average number. Sometimes I do up to 10K req/seq. That consumes some noticeable amount of CPU, but nothing special and the bottleneck seems to be not here.

And I think the main issue is the "Latency" chart. Because my average latency is about 3.5ms. It's like 10x faster. They cannot be so slow.

So I guess the real problem is that the author measures the network. Not the service. So a link to Alchemy happens to be slower than a link to his own server. And the network is the main bottleneck in both cases, because you can easily do 10x of those numbers with a good network.