Oh, it works, all right. Check out the two images. The first has a displacement on the bark of .005 meters - relatively smooth. The second has a displacement of .5 meters.
<edit>
The third has displacement set to .5 meters, and uses the color map as the displacement source, instead of the grayscale version. The difference between the second and third doesn't really look that profound, but with reasonable displacement values the color-based displacement looked blurry, less distinct.
</edit>
And it's really displacement mapped, if I'm not mistake. It's not applied as a bump-map. And the grayscale thing makes sense, if you think about it - that way all the components of a displacement vector are the same magnitude, unaffected by the color. If the unequality of the color components carries over to the displacement vector components, you get ... inconsistent? ... results.
The tree in the image is one of two new smaller palms I just did. The scene I'm working on is heavily displaced and it's difficult to get populations to sit correctly, and if I try to vary the scale it affects the offset I need to get the pops right, so smaller trees disappear into the terrain. So.. I made smaller models I can mix with the big ones, using a fixed offset, and it looks pretty good. Give me the word and I'll send 'em over.
<edit>
That figures. About an hour after I finished putting the final touches on the smaller palm models I found the problem with the terrain displacement/compute terrain discrepancy. Now stuff sits exactly on the terrain, and I can use the built-in size variation feature, largely eliminating the need for the smaller models to begin with.
</edit>