IBU estimation is a black art. As noted, there are many different formulas for doing it that can vary wildly in their predictions--Tinseth, Rager, Mosher, Daniels, and Garetz are the most common, though there are others. Basic Brewing Radio did a show on this--they used several different formulas to guess, then actually had the real IBUs measured in a lab.
One recipe varied from 18 IBUs to 74 IBUs as estimates (actual: 34).
In some cases, the spread of the formulas didn't even contain the actual answer--e.g. a hefeweizen where estimates ranged from 12 IBUs to 17, but actually measured just 7 IBUs.
My personal takeaway was:
1. Pick one formula and use it the same way consistently; that way you'll have a consistent baseline that you'll learn how it comes out flavor-wise (I chose Tinseth, as it seemed most accurate to me); and
2. Only expect accuracy to about +/-30% in the estimates*. If the prediction is 50 IBUs, you can be reasonably confident that it's in the 35-65 IBU range, but don't expect it to be spot-on.
*Or even worse for unusual beers (e.g. anything with wheat/rye, partial boils, unusual hopping schedules, crazy yeast, etc)
On deck: Little Bo Pils, Bretta Off Dead (Brett pale)
Secondary: Oude Bruin, Red Sky at Morning (Sour brown ale)
On tap: Saison Duphunk (sour), Amarillo Slim (IPA), Earl White (ginger/bergamot wit)
Bottled: Number 8 (Belgian Strong Dark Ale), Eternale (Barleywine), Ancho Villa (Ancho/pasilla/chocolate/cinnamon RIS), Oak smoked porter (1/2 maple bourbon oaked, 1/2 apple brandy oaked)