The most obvious culprit with Fluent is that it is allowing the flow to be turbulent while your model assumes laminar flow. However, this would produce an error in the opposite direction. You'd expect a higher pressure drop in the turbulent case. In other words, if Fluent was modeling the flow as turbulent, you should see a higher calculated pressure drop, and your answer would tend to be higher than theory (if the theory is correct).
On the other hand, it is possible that, for a U-bend in the pipe, there would be additional losses associated with the bend that aren't accounted for in the Darcy-Weisbach equation. However, this again would tend to mean that your predictions get worse than they already are, not better. In other words, if all of your parameters in the Darcy-Weisbach equation were correct and the problem was leaving out losses associated with the bend, then you would expect that theory again is predicting less than Fluent (assuming your Fluent is correct).
The next best bet is that your friction factor is not estimated properly for the geometry that you put into Fluent or that something was wrong with your Fluent simulation. I don't know what to tell you there. Maybe your estimated your Reynolds number incorrectly or turned the wrong bells and whistles in Fluent.
