From the README:
  I propose this delta generator as a replacement for "diff" in applications
  where the insert/delete delta is not required.  Since the copy/insert delta
  is easier to compute, it does not have to make the reduction in input size
  by breaking the file into lines.  The delta generator works well on binary
  files.

  The next question is whether it is actually desirable to compute deltas
  between binary files.  The answer is certainly yes, even though some binary
  file formats will not have a great degree of similarity between versions
  which were generated between minor modifications to their sources.  First, I
  have evidence that some file formats (notably FrameMaker documents) work
  very well.  Machine-dependant object files and executables don't work very
  well, but it is still worth doing.  The reason it is still worthwhile is
  that compression takes longer than finding these deltas, so any space
  savings the delta generator produces it very likely to reduce the total
  archival time.  Even if the delta generator saves no space, the total time
  well end up less than twice the time of compression.  I will include some
  measurements at the end of this writing.

  Please send any comments, questions, thoughts, or interest to me,
  jmacd@cs.berkeley.edu.

Eric. <erich@FreeBSD.org>