LoD is a technique for decimating meshes based on distance without distorting UV's, similar to mip-mapping with textures. The idea is that a geometrically complex object is already obscure past X distance (or X1, X2, X3, etc. for various levels), so we either use an algorithm to generalize the surface and UV's (that's texture coordinates) or store pre-decimated meshes that get displayed at the specified visual ranges.
If you don't see a difference you might not be looking at distant enough objects, or the function is doing its job really well. Since many EQ models are extremely low-poly by today's standards, it stands to be more noticeable in newer zones with fancier geometry and textures.
http://habibs.files.wordpress.com/20..._of_detail.jpg
Edit: What he said ^^