After deriving the equation, I solicited other brewers' data. I don't have the numbers on me, but six or seven people contributed 70 or so data points, and the standard deviation for those was 0.0013 - a little more than one "gravity point". So statistically speaking, you could expect the correlation to be within one point 56% of the time, two points 88% of the time, and four points >99% of the time.
I will say that in my own experience the standard deviation is a bit smaller than the overall sample. So when using consistently and thoroughly calibrated instruments, the calculation is about as precise, or even a bit more so, than a cheap hydrometer.