Besides a whole host of logistical concerns that AugustDay points out, there are a few hard theoretical bounds that preclude your first point about using these machines to perfectly simulate solar systems. These all basically hinge on one fact: the universe is very big, and there's lots of stuff in it. If you ever work in computational physics, you'll find a very large portion of the work comes down to figuring out how much you can simplify your model and not get a totally garbage result, because there's just too much stuff to simulate everything exactly.
To give a more concrete answer, let's forget about GUT level simulations and assume we just want to simulate plain old Newtonian mechanics perfectly, which is far simpler. Well, to do this, note that as a rough estimate, there are about $10^{57}$ atoms in the solar system, and for each one say 1000 bits is enough information to accurately describe it. This means that each time step will require on the order of $10^{60}$ bit operations. I'd say that a time step of at less than $10^{-15}$ seconds is necessary, as humans have measured time scales around this range. So that gives a value of at least $10^{75}$ bit operations per second of simulation time. That's astronomical-- even if you have a star at your disposal! And keep in mind this is a severe underestimate, since it doesn't take into account quantum mechanics, relativity, or even electromagnetic fields.
To get an idea of how impossibly large this number is, I'll introduce you to Landauer's Principle, which states that in irreversible computing (ie all computing we've done so far), any irreversible manipulation of information requires at least
$$kTln(2)$$
joules of energy. In space, the lowest we can get T is about 3K, which is the temperature of cosmic background radiation. You can make colder environments, but this requires additional energy for refrigeration, which would take away from the energy available for computation. Combining with the sun's luminosity of about $4*10^{26} J/s$, we find the information theoretic limit of about
$$10^{49}operations/sec$$
for what our sun could produce. That's nothing compared to what we need, even assuming our computer works at the maximum possible efficiency (which none of our current computers even come close to). In fact, the computer using our sun wouldn't even be able to get through a single time step before the sun died!
And the situation gets even more bleak if we consider GUT level calculations like you say. In that case, the formulation of QFT requires us to create an extremely fine grid to which we assign several fields. Humans have probed to at least $10^{-18}m$, so if we want the simulation to convince them, we would need a grid with a resolution at least as fine as this. But this requires $10^{54}$ cells per cubic meter, so even simulating a single cubic meter would be pushing the limits of our computer.
Now, it might be possible to wave these concerns away by saying something along the lines of "perhaps they've utilized reversible computing", but really you can magic away any scientific objection by saying science has advanced to unforeseen frontiers. I just wanted to give you an idea of how difficult it would be to implement your plan according to modern understanding, even in an idealized world.
EDIT- I just realized I was thinking of years instead of seconds when I said the sun would die before a time step occurred. So in reality, the sun could perform about $10^6$ of the time steps I described. To be fair, my estimate of $10^{60}$ operations per time step is a criminal underestimate, since it doesn't take into account the operations needed to compare each particle to every other particle and do arithmetic. A more accurate number taking this into account would probably be around $10^{120}$ operations per time step.
