Here's the math answer. Please know that I picked the example numbers because they're easy to work with; I would expect much lower variance from actual powder measures:
It's going to depend on your specific powder throwing mechanisms, and whether the variance of a given device is a percentage of the overall measured amount, or by an absolute weight.
If your thrower varies on absolute weight (so by, say, +/- 1.75 gr., no matter your target weight), then by throwing twice you've doubled your maximum error as Dusty points out. If your measure varies on a percentage of target weight (say, +/-10%), then it's not that you wind up with smaller errors (since you still have the same total potential error margin), but rather that you can wind up with cancellation of errors.
I setup a sample in Excel, assuming a +/-10% margin of error for a given throw, and assuming that heavy and light throws occurred with equal frequency and at random. I then used the Excel random number generator to give me a set of 600 values from 0 to 1; half were multiplied by 0.1 and the other half by -0.1 to determine the percentage of error for each throw. Those errors were then applied to a charge weight of either 8.75 (half throw), or 17.5 (full throw). The half throw charges were then randomly sorted to distribute the heavy and light charges, before those charges were added together to derive the total charge list. Selected stats for the charge lists are below.
Half Charge
StdDev 0.74621
Extreme Spread 3.02
Max Charge 19.06
Min Charge 16.04
Avg Charge 17.56
Median Charge 17.56
Full Charge
StdDev 1.002899
Extreme Spread 3.47
Max Charge 19.24
Min Charge 15.77
Avg Charge 17.46
Median Charge 17.51
So if you have a thrower that varies based on percentage of target weight, it would appear that half-throwing your charges may produce more consistent loads, with a less extreme spread across the sample. Whether or not that extra consistency makes a difference over, say, weighing each load, is going to vary on a case-by-case basis. It's something you would need to test and verify for each measure/powder combo you want to use.
The main risk of this is the need to activate the lever twice, and thus adding double the human intervention and double the risk of human error. I say this as someone who tracks and works to mitigate medication errors in healthcare: each touchpoint is an opportunity to introduce error into the system, and so should be examined for necessity and potential alternatives. In addition to the safety aspect there's also the time aspect: the faster I can load quality, safe ammo, the better.
My preferred system, both from a time and safety standpoint, is to weigh every charge on my GemPro. I drop most of a charge (within 0.2 gr., usually) with my Lee thrower, then trickle up into a pan on the GemPro. +/- 0.06 gr. is acceptable for me; anything that's too heavy just gets dumped back into either the thrower or the trickler, then gets re-thrown.