In reply to Chris Reid:
As mentioned by others, ENSA has tests that suggest the figure 8 is as good or better than the ordinary EDK. . This result doesn't agree with tests done by others. It also seems that when they use ropes of unequal diameter and an EDK, they fail to use the best configuration for roll prevention. A potential issue is that the ENSA tests seem to be conducted at a relatively high constant pull rate, something that doesn't seem to me to model any climbing situation.
One thing you can see in the ENSA tests, already remarked on by Moyers in the original set of tests, is that the EDK "rolls" whereas the figure-8 "capsizes." The difference is that the figure-8 rather suddenly pops into a new configuration, chewing through a lot of the tail in the process. This behavior by itself would, from my perspective, motivate forearm-length tails.
For those of us who have been rappelling for many years without knotting the ends of our ropes, the compromise idea of not tossing strands with the snagging equivalent of a nut on the end but still protecting against going off the ends if the landing is sketchy seems like a good idea. It also allows the climber to tie both strands together with one knot, something that frequently works out very badly if you try it from the outset of the rappel.