The effect doesn't have to be expensive, and my original implementation was fast on millenium era machines. The goal is preserve the context of the background, while keeping foreground data readable, in a way that is natural to our human visual system. If you are doing it properly you also shift the brightness values into a more limited range to diminish the contrast and keep the tonal values away from the chosen foreground text color (this is also cheap). Done properly it is visually pleasing with virtually no effect on readability.
People have coded some boneheaded imitations along the way though. They don't add the blur, or they don't adjust the brightness curve, or they make the radius much too big, or they compute some over exact Gaussian blur that is too slow.
It's the nature of blur that it doesn't have to be exact to be visually pleasing and convincing.
I don't think it's the nature of progress so much as it is laziness. Most developers (myself included) don't worry much about optimization until the UX performance is unacceptable.
I sometimes wonder what the world could be like if we just froze hardware for 5 years and put all of our focus on performance optimization.