For any AA system to work, it needs to know more about the image than is actually visible. Applying any sort of blur will affect vertical and horizontal edges in the same way as jaggy diagonal ones, so the result is just a blur, without an AA effect. The AA system needs to 'know' that a jaggy line is supposed to be diagonal, and apply AA 'blurring' accordingly, and the only way it can know that is if it has access to more detailed information about the image than you can see.
The AA applied at the time of rendering has access to the mathematical definition of each edge, so AA can be applied with a high degree of accuracy. Edges that are supposed to be vertical or horizontal are left alone, whereas V & H edges that are actually making up a diagonal line can be anti-aliased to provide a clear, jag-free image.
The only post-render AA effect that I have found that has any degree of accuracy is to apply a light Gaussian blur to a render with at least double the resolution in both axes (twice the width and twice the height) and then resampling down to the required resolution. For example, if producing a PAL animation with 720x576px, you would render at 1440x1152px, apply the Gaussian blur, and resample down to 720x576px. In this way you are still going from a 'more information' state to a 'less information' state, so some degree of AA is possible.
The effect is not as good as true AA, and the technique isn't much use if you have already rendered your image. Although the larger render will obviously be slower to produce, it will still be significantly quicker than rendering at a lower resolution with AA on, at least out of Anim8or's renderers.
