Sorry for double posting, I'm going to reply partially to my own question 😉 Please correct me if I'm wrong.
By looking at the starling code, using embedded fonts is pretty similar to what I already did, i.e. putting my text into a regular Flash textfield, then rendering this textfield into a texture. So if the text is too long, it will be truncated anyway (moreover with the Starling textfield, we have to anticipate the height of the rendered text, else it will be cropped, at least with the embedded fonts).
With bitmap fonts, every char seems to be a seperate object, then everything is flattened. I suppose that flattening results in one single texture? In that case, we have the same constraint.
All in all, it seems that we can't have a Textfield higher than 2048 pixels on most devices, since it always ends into a texture? Or did I miss something? 🙂
I think I'll continue with the method I was using of rendering my own Flash textfield into a texture, splitting the textures if the text is longer than 2048 pixels.