Evidence Based Scheduling

This post is a bit old, but still one of my favourites. Way too many projects are planned this way: http://twitpic.com/tsb9f

Has anyone tried Joel's approach in their own project? Would love to hear how well it works in real life.

In a current project, we're again struggeling with an unrealistic schedule, combined with an 'agile', iterative approach where all product features are fixed in a 3000 pages long specification.