This is true, but, as you said is simply a technical option... that don´t exist out of the virtual world.
I mean that box-modelling is based in the natural way of drawing, first you do the boxes, then the bigger areas, and finally, the small ones, a natural fractal process.
I won't dicuss this any further beyond this post, since it's way off topic and could become sort of philosophical, but:
The box is made up of polys. You can start with it as a base right away, but in case of drawing, you can't do this. You have to draw the box first.
And that box is made up of lines which are made up of strokes which are more equivalent to polys than to a box.
If you're a sculptor and start with a block of marble or something, then I see how you always "start with a box". But if you're a drawer or painter, there's absolutely no need to do so and even if you do, you first have to draw it.
Of course, you're refering to constant refinement of a given layout. This is true for all arts.
If you model something in 3D, there are some techniques that are typically thought of as "box-modelling" techniques, like subdividing a side of the box and some that are rather thought of as "poly-by-poly-modelling", like extruding a free edge into a ridge over mutliple extrusion steps.
But the reality is, you can only truely differentiate between the two in the beginning. Do you start with nothing (an empty canvas or maybe an edge that, unless you choose to draw mesh grids later, wouldn't even be visible in a render) or do you start with a box (a marble block that you can sculpt into the shape you want).
At some point, and that can come pretty early, if you're an artist with a decent amount of experience, you'll use techniques of either "camp".
Which techniques you think of as superior might heavily depend on the 3D package you use or its supported featureset.
For instance, in Blender3D, not too long ago you didn't have NGons, so working the Box-modelling way was highly limiting. Most users quickly learned that it's just way easier to go for Poly-by-Poly. However, now that NGons are supported and we see a couple of new, really powerful modelling function features (sure to increase quickly), it's much less limiting to do Box-modelling and more and more users will use that.
Even before NGons, there was quite an increase in Box-modellers with the implementation of a sculpt-mode that, by now, is pretty powerful aswell...
Ok, I'm done. I kinda went on a tangent there. All I'm saying is, that even "physical" art can start either way.