I'm not a physicist, but I would guess that real water is more 3d (or perhaps it inhabits some N-dimensional space filling it with sub-atomic particles, I don't know, I'm not a physicist).  More to the point:

In most cases we cheat as much as we can, to gain performance.  If you can create the impression of something (like water) in the viewer, than it doesn't really matter whether the underlying physics you employed is accurate, only that the end result is what you want.  When programs try too hard to emulate real world physics, they can end up suffering performance penalties.  

Sometimes, the anomalies we introduce by violating real world physics turn out to be delightful.  As procedural artists, we have the freedom to rewrite the laws of physics, or to follow them, as we like.  Isn't that wonderful?

jburn already pointed out the crossover between science and artistic interpretation.

In any way, you need to set up a (mathematical or physical) model, and this depends on your aims and also the size regime you want to model (from a single water molecule to the interaction of several molecules up to the macroscopic regime that describes the liquid (it may also be a solid or a gas) that starts beyond a few nanometers up to earth scale). Of course, the closer your model is to a physical description, the more realistic your result will be. But this also means that the equations of your model will become more complex and more equations have to solved simultaneously, resulting in the performance penalty jburn has described.

Is your aim

1. an animation - 2D is ok. Size regime may be molecular up to macroscopic.

2. an artistic rendered scene, e.g. a water surface that mirrors, transmits or absorbs light? Needs some surface modeling (the model can be 2D that is then projected to 3D) and then some ray rendering software (usually working in 3D).

3. a simulation of flows (motions of natural waters, oceans, simulation of mass transport in pipes, channels, ...). Needs a fluid simulation software to solve the Navier-Stokes equations numerically and is usually done in 3D, but for simple cases (e.g. centrosymmetric pipes ) can be done in 2D.