10000 players in one area is pretty hard on its own. The scaling is quadratic since each player receives the updates of all players.

Simulation is irrelevant in terms of performance because it is a fixed cost that is shared across all players.

I don't think so, as the unit of simulation are chunks, and only chunks near players are stimulated - at least in classic Minecraft. I.e. if all players are in one chunk only one plus the surrounding ones are stimulated, but if all players are spread far apart, you have the same simulation workload for each player.

Or is this different in Hyperion?