Thanks. That is really intriguing. Did you look at those two files as plain text as well as via ghostscript or similar? I'm not familiar enough with postscript to understand how the 5 characters changed in the binary bit at the top can cause two otherwise identical documents to appear so different when formatted? Kind of reenforces my distaste for non-plain text communications mediums.

Even though it is rare I try to avoid programming with the view "that probably won't ever cause problems".

Agreed. The 'problem' with the MD5 hash, and all other hashes for that matter, are applications that use them under the assumption that either clashes cannot happen, or are so rare that there is no need to verify them. Especially for security/cryptography applications.

The assumption that any digest/hash function that can represent any size document of file with a short, fixed length 'unique' signature is mathematically impossible (a bit like infinite lossless compression :), and any security application that relies on that in just plain broken.

About the best you can do is compute two more different digests of the document which should make it much, much harder to generate two disperate, but meaningful documents that produce the same digests through the different hash functions.