Are there *that* many retries and collisions during a large high-bandwidth file download?

You'd think that downloading a single large file in a constant stream would be the thing that was simplest and which resulted in the fewest problems.