The thing is, while it might mandate efficiency, what they made was an un optimized 8 bit VM synthesizing 16bit arithmetic running on 64 bit hardware. It's not efficient at all, it's just that the constraints they gave themselves leaves enormous amount of power on the table.
you will probably be interested in the efficiency measurements i made in the thread linked from https://news.ycombinator.com/item?id=41132868. the standard sdl uxn implementation is less inefficient for things like text editing than you'd expect from the facts you mention, and i suspect uxn11 may actually be pretty decent
SDL2 has been built as a fast whole/partial screen refreshing platform for textures/images, no wonder if it's that slow mimicking XFT/Cairo's job on fonts/text. Just look at the Emacs editor for Common Lisp vs the SDL build of Lem.