Theo, you can disagree all you want, but you'll still be wrong.
First of all, the UO Pseudo-random Number Generator is not a "fair" (in statistical terms) number generator. This is shown by simply crafting a few thousand items and looking at the evidence. For example, the chance of crafting 10 normal items in a row at 90%+ exceptional should be astronomical, something that happens maybe once a year to a person, but I get it over once a week. I've had the same sort of streak happen with the number generator, producing 10 exceptional items in a row at 10% chance of exceptional.
You graph this out, you get something that has nothing remotely resembling a (statistically) normal curve's standard deviation. This is especially true when you consider each failure is an automatic 50% loss. This isn't a heads or tails proposition; Order DOES matter, for results. On an experiment where one is trying for at least 1 heads out of 3 tries, the experiment doesn't stop if one head is gained, nor does it treat a failure on the first, and success on the second as only "Half a head". I'm oversimplifying my argument a bit, but the math involved is not the basic Statistics & probability type stuff one learns in a few weeks of High School Advanced Math, but more in line with stuff from second-term college Statistics courses.
So, as someone with a background in math (was a math teaching major - completed all my math classes but the senior level ones, before my health failed), I can confidently say that based on the faults in the UO system of generation "random" number, all arguments trying to use normal distributions to define smelting rates being the same over time, regardless of method, are bunk.
Also, because of that wild variablilty within the system (failing 5 times in a row on something with a 90% success chance shouldn't be happening weekly on a miner with only 1000 experiments/smelts a week), the "Large Sample Sizes" needed for a fair set of experiments are on a scale normally seen only by script mining operations like the ones whose houses are burning. The abnormal curve the flawed RNG is producing requires absurdly large samples, especially compared to the number of experiments in each, for the math to work out.
Smelting in smaller groups of ore guarantees that one won't get the maximum return, but at the same token, it makes it much more likely for the results to resemble a point along a normal distribution, and all but eliminates the flawed RNG's tendency to give you less than 6.25% of your ore that the numbers say you should get 100% of 90% of the time.
If anything, the math behind the full expression of the system more resembles the math determining how much retail sales it takes at a set profit margin to cover the losses of the wholesale value of an item that was destroyed or stolen - but even that is nowhere near as complicated as trying to make sense of the random number distribution of UO.
