Is there really less information required to represent text than data? It just seems easier to encode ascii text because we've settled on an encoding schema that optimizes for the english alphabet. The example of the twitter icon using 2000 bytes is only because the author decided to use a png. Using the font-awesome typeface, the twitter icon is just 2 bytes. And I can whip up in 5 minutes a typeface where the twitter logo uses 1 byte. We can come up with an encoding format where dingbats and logos take up 1 byte and the english alphabet uses 2TB. It'd be a useless encoding for practical purposes, but goes against the idea that text is inherently more informationally compact than images.