Personally I really doubt (2). I work in theoretical condensed matter physics, and it's awfully hard to accurately simulate anything with more than a very small number of atoms. Sure, computers will continue to get faster for a while, but these problems scale very badly with the size of the system, and we're not all many orders of magnitude away from the fundamental limits of computation. I suspect that even if you were to convert the whole mass of the solar system into one big computer, you probably still couldn't run a decent simulation of any system on the scale of 10^23 atoms.
It's possible you might be able to run a somewhat dodgy simulation which doesn't work on the atomic scale unless it has to (e.g. when your simulated person decides to do some atomic-scale experiments, it would suddenly launch a proper simulation, and otherwise it could get away with a neuron-level simulation). But I still don't think that we'll ever get the computing power needed to make these kinds of simulations trivially easy.
It may be possible that each universe will simulate a substantially simpler universe; it would of course be impossible to tell whether you universe has "lower resolution" from the inside, cause you'd have no frame of reference.
Perhaps our universe's resolution bottoms out at the quantum level, where a hack saves processing at the lowest level by using probabilities and a lazy evaluation scheme instead of deterministic calculations.
One possible solution is to elongate the time-steps; if we take a million years to simulate one second of simulation time you could get by with a lot less computing power.
If you do that, and load some information from a slow storage device (hard disk), calculate on it for a hundred years, then write it back, load some more, repeat, until you have computed "one planck time" accross your entire simulated universe...
... at what point in the loading, crunching, saving, does the information inside feel concious and self aware 'in realtime'?
Since this could apply to your life being run on a different substrate, with no actual effect on your perception, I don't see how this is a useful question.
The problem is that the scales involved are so ridiculously large that it could easily take 1 billion years to simulate one second. Even with 1 million years per second, simulation of a thousand years would require something like the age of the universe to compute. In that case, these "beings" could not be our descendants since they live in a universe with different laws.
Actually, at a million years per second, the age of the universe (so far) would go by in five simulated hours. I don't want to speculate on how long we have until heat death, but I doubt it's long enough to simulate a single human lifespan at a million years per second.
* Assume humans are the target of the simulation - our senses are fairly constrained (eg our focal point is only a tiny area where hi resolution is required etc...)
* What about non classical computing (quantum computing) - it may end up being nothing, but it seems to have some interesting promise in going beyond the raw computation power of atoms at least.
You're assuming that we'll be using the same approaches (e.g. molecular dynamics on classical computers) to do the simulation. We don't know what methods and techniques will be available in the future, so we can't confidently claim that it'll forever be out of reach. I bet there's some room for improvement over the current crude techniques.
I'm not saying that we'll have arbitrarily powerful computations, but that there's a lot more room for improvement. Like Feynman said, "there's plenty of room at the bottom".
It's possible you might be able to run a somewhat dodgy simulation which doesn't work on the atomic scale unless it has to (e.g. when your simulated person decides to do some atomic-scale experiments, it would suddenly launch a proper simulation, and otherwise it could get away with a neuron-level simulation). But I still don't think that we'll ever get the computing power needed to make these kinds of simulations trivially easy.