No, you're not doing anything wrong. There are image renderers for graphical views for numeric and binary, but for multistate there is only the basic renderer. We will add a such a thing for the next release.
Indeed it is a wishlist, and your point is valid. It's true though: the background and all of the point view images would need to be scaled, and the effects on the (un-scalable) text renderings would need to be taken into account. (Sure, text can be scaled, but only in a very quantized way compared with the granularity of image scaling.)
Overall it is a tricky thing to make work properly, but still possible.