I addressed this issue in a paper a few years ago: http://www.guyboulet.net/site/docs/Rapid_prototyping.pdf.
The main problem with ADDIE is that it is a cascading model where each phase depends on the results of the preceding phase, assuming that the deliverable of the preceding phase are complete and accurate.
What if an assumption made in the analysis phase is incorrect and this leads to incorrectly targeted design and thus to development and implementation of a solution that aims at the wrong target? You might only find it out at the evaluation phase, at the end of the project.
Because of that, I believe that iterative models such as SAM or Rapid Prototyping are more efficient for elearning development as they offer early detection and correction of problems.