A player might be connected to several servers.
If you divide the population of an area in multiple servers, but still connect each player to each server, that won't remove the complexity, you are just replicating the areas and their population. If you want to split the population you have to communicate the severs between them.
instead of a few complex interactions I want many simple interactions
If the server is not authoritative, it doesn't matter how simple or complex the interactions are, since these are calculated in the client side. But if you don't want players to cheat you need to calculate most of the interaction from the server side. Even the simplest approach doesn't remove the requirement to transmit and process a huge amount of data in a small frame of time. And the problem is that you have to return the result of every other player in the same area to every player.
The main theory behind cloud is that you can easily add and remove servers in your system while it is running.
I think that theory is related to clients that don't interact/see all of each at the same time in milliseconds. Maybe for 5 people working on the same document and updating changes every few seconds? But when there is full interaction, there is probably a decay in the capacity achieved by stacking servers. If you manage to optimize this to the point of making it linear, then you will be spending a linear amount of money. That is without including the bandwidth, infrastructure and maintenance/managerial personnel of the farm. Otherwise, the amount spent will be exponential until reaching a point when the stack of servers won't be effective. You could prolongate the curve by creating a hierarchy of severs, but the exponential cost will still persist.
It's an interesting problem