The datasheet is based off of real world testing (by most manufactures anyway) so I would trust it. After you come up with a model (on paper or via a computational model) and it's different, then ask why the model differs from the real world.
Sometimes you need a more accurate model, it looks like you haven't modeled the parasitics (which might not make a difference in this model because they are usually small and affect ~10kHz and above). Wires, resistors, capacitors, inductors all have capacitance, resistance and inductance that may need to be modeled.
The second problem is manufacturers do not model all of the small order effects of a spice model. Sometimes they model very little and only match bandwidth, output and input impedance. Some op amp models capture 90% of the effects. Almost none of the spice models are replicas of the original circuit, because then they'd be giving away their design.