It's not my field of expertise, but my very-slightly-educated guess is that the accuracy difference between 2x500lb and 2x1000lb load cells will be so infinitesimally small as to be completely outweighed (pun fully intended again) by inaccuracies caused by general installation conditions, like termination resistance, length of cable run, interference, alignment of the load cells, accuracy of the calibration weights, whether the load cells are 4-wire or 6-wire, accuracy of your A/D converters if there's any sort of 4-20mA or 0-10V analog output/input involved, and so on. And, of course, you'd have to consider whether repeatedly overloading one of your load cells will eventually lead to more inaccuracies.
I mean, if you were saying "should I use 750lb load cells or 75,000lb load cells", then sure, I should imagine there would be a loss of accuracy if you went the 75,000lb route. But 500lb to 1000lb? Not really.
What you're really changing is not the accuracy, but the resolution. A load cell has a mV/V rating, typically in the range of 1-3mV/V. Let's consider a 500lb load cell with a rating of 2mV/V, connected to a weight transmitter with an excitation voltage of 10V (which is fairly common). A rating of 2mV/V means that the signal output by the load cell will vary by 2mV for every V of excitation voltage, across the full rated load. So, the difference between no load and 500lb is 2(mV/V) * 10(Excitation V) = 20mV. This means each lb gives a difference of 0.04mV.
If you consider the same setup but with a 1000lb load cell, it's now only 50% loaded at 500lb, so the difference between no load and 500lb is 2(mV/V) * 0.5 (50% load) * 10 (Excitation V) = 10mV, so each lb gives a difference of 0.02mV
Now consider a 50,000lb model. At 500lb it's only 1% loaded, so the difference between no load and 500lb is 2(mV/V) * 0.01 (1% load) * 10 (Excitation V) = 0.2mV. This means each lb gives a difference of 0.0004mV.
From the perspective of your weight transmitter, it's just converting a mV signal to a digital signal internally, which is then displayed on the screen and output via 4-20mA or 0-10V or some kind of fieldbus. It has an A/D converter to convert this mV signal to an internal number, and that A/D converter will have a certain resolution. Let's say that the resolution is 0.0005mV ( a figure that I pulled out of thin air for the purposes of illustration).
So, for your 500lb load cells, where each lb = 0.04mV, your maximum resolution - i.e. the smallest change in weight that you can measure - is 0.0125lb.
For your 1000lb load cells, where each lb is 0.02mV, your maximum resolution is 0.025lb.
For your 50,000lb load cells, where each lb is 0.0004mV, the smallest change in weight that you can measure is 1.25lb.
Now in practice this looks like the 50,000lb load cells are inaccurate, because you can put an extra pound of weight on the scale and it won't change. But the fact is they're accurate, they're just not as precise as the 500lb or 1000lb in that range.
Again, comparing between the 500lb and 1000lb options, they will both be as accurate as your installation conditions allow, and the inherent difference between them in accuracy will almost certainly be so negligible as to not make any practical difference. If the resolution is of critical importance to you, then you would have to carefully select a solution based not just on the rating of the load cells, but also on the type and resolution of your weight transmitter.
Again, I'm not an expert on this. I've just messed about with load cells enough to absorb a bit of information about how they work. Take all of the above with a grain of salt.
One final anecdote about how equipment selection is only one very small part of a weighing solution. I was commissioning a batch weighing system where batch weights ranged from 150g to 30kg. Acceptance criteria was +/- 2%. +/- 2% on 30kg is over half a kilo, so we had no problems whatsoever with the larger batches, but +/- 2% on 150g is 3g.
We struggled badly to get the required accuracy on the 150g batches - we'd get two in a row, and then the next three would fail. Until I noticed something. Across the room, about 20m away, there was a fan on the wall to keep the operators cool. It oscillated back and forth slowly. If it was pointing at our equipment when the 150g batch when through, it would fail every time. I turned the fan off, and we put through nine 150g batches in a row to within 1g. On the tenth batch, a forklift drove past and the batch failed again.
All that goes to say - weighing is an inherently complicated task, and when you look at all the variables in play, the difference between a 500lb load cell and a 1000lb load cell is often completely academic from the perspective of "how accurate will my entire solution be"