Yes, one of the big differences is that 'tar' lets you break up larger 
filesystems into smaller, more manageable DLEs, as well as excluding 
specific files or file patterns.

Be sure that you use 'tar' version 1.13.25 -- nothing older, and nothing 
newer, until you hear that someone has certified it to be compatible 
with Amanda.  (The latest version, 1.13.92, causes indexing not to 
work.)

Also, I've found that 'tar' seems to take longer in the estimate phase 
than do 'dump' (and 'vxdump'), but for all I know, that could be due to 
local influences (e.g., disk traffic).  Or has anyone else also found it 
to be the case that 'tar' estimates take longer than those from 'dump'?

-- Mark