Which means that at any given real-world moment, some parts of the simulation are at timeslice T, some parts are at timeslice T+1, and so forth.

If the six-hours is to avoid too much time-skew, it's a hack and one would expect better from simulation-builders.

There are plenty of ways to efficiently calculate differing time-space regions, using lazy evaluation or equivalents thereof. For example, the famous Hashlife algorithm for Conway's Game of Life does exactly that - different regions can be billions or trillions of generations apart thanks to memoization. Lazy evaluation proper allows weird techniques like the reverse state monad or circular programming (aka time-traveling).