This is only true if you are trying to extract absolute maximum brightness from a LED. If you have "headroom" and can operate the LED closer to its linear region, then you can use multiplexing to increase your efficiency.

If you reduce the intensity by multiplexing, you reduce power - yes. How come you do not reduce power if you reduce the intensity in non-multiplexed drives???

Forb illustration grabbing a datasheet at random, the lite-on LTJ-811, the datasheet can be found at http://www.digikey.com/ (the absolute link 'dies', you will need to enter LTJ-811 in the window and go through a few steps) it should be clear that the non-linear region only apply when running the display at so low an intensity that power consumption is no concern anyhow.

Erik