The online earnings calculator is pretty flawed. In your example of holding 3TB vs 5TB of data, it’s fairly reasonable to conclude that if you hold 66% more data, you’re likely also seeing 66% more egress traffic on average. After all assuming the chance of download of each piece is about average, holding more pieces would simply lead to more downloads taking place. (You should probably also not use the current test traffic as a representative pattern of downloads in the future)
Each transfer is already handled in a separate thread by the node, so it’s always dealing with separate simultaneous transfers, there is no additional network overhead for spreading that same amount of traffic across multiple nodes. There may be a slight increase in CPU use and RAM, but I think most systems can deal with that perfectly fine. And either way the theoretical tiny bit of overhead it would add is no way going to make up for the 33% loss of revenue for sacrificing an entire disk.
That’s fine, I’m not here to convince you. Just sharing my point of view based on what I know, you’re free to make your own decisions.