A more boring answer is probably risk management. You don't want to be the radical that introduced a new innovative interview process and produced false positives for your company.
It's like how everybody knows IBM sucks but nobody gets fired for choosing IBM (TM)
If you want change, it has to start at the top of the management
One has to admit, there's also a certain elegance to the l33tcode thing. It's all about easy-to-survey "code katas" taken to an extreme level, which is pretty much what you'd want if your problem was to cheaply select the most likely capable 1% or so out of some huge amount of applicants. But most real-world code problems look nothing like those clean whiteboard examples.
It's like how everybody knows IBM sucks but nobody gets fired for choosing IBM (TM)
If you want change, it has to start at the top of the management