I see what you mean, good catch.
In the context in the article the author is talking about taking the dimensions of use independently, of number of messages, payload size (together which combine to throughput) and number of nodes. I thought it was that for the same number of nodes, hashgraph can scale in throughput, and for the same amount of throughput, it can scale in number of nodes (that is, holding each variable and scaling the other) but perhaps my interpretation was incorrect. Thanks for that, now I'm not sure.
I'll need to do more reading actually because there's not quite enough here. I'll get back to you on that.