It would depend on a high number of factors, wouldn't it? You'd need to consider the possibilities of solipsism and if everything is always loaded in, or if instead things are only in existence when looked at

For factors, choose the way that minimizes the size of the host machine, while keeping the result meaningful, e.g. each living human is a distinct interactive member.

Maybe the speed of light is the equivalent to how long it takes to load from storage

That would imply the computer we are simulated on is isomorphic to the 3D universe it is simulating, i.e. that if A and B are close in the simulation, the storage for A and B are physically close in the simulator, and that the storage moves as the objects move. Otherwise the speed of light would vary from a place to another.

None of that matters if the simulation only has access to simulated time--e.g. a vm with access to a clocksource but not a 'real' time one.

Computers have to run even when we're not looking at them, at the very least.

Unless.. it's only there when you're watching!

As soon as you turn your head, the garbage collector is right there to sweep it up.

I think that it could be lossily compressed, much like a bloom filter. It dedups as required but then some of our experiences/memories get changed/mutated/compressed. The difference of course is that unlike a bloom filter the reference is lost so there's no 'error term' to speak of.

I was expecting to see the area of the universe / Planck units or something to that effect.

When trying to wrap my mind around it, I feel like simulation implies a sequence of events like a movie, but to an outside observer the universe would be more like an image.

I can't find it right now, but I recall some paper by Ed Fredkin that claimed that measuring fundamental physical constants indicated a 256-bit register could work.

I think the paper was in an IBM in-house technical journal.

Bits? ;-))) Gödel's first incompleteness theorem says no bits where we are going.