As was pointed out by a commenter earlier when Bruce Schneier posted this.
This whole hypothetical is moot and has already been attempted for DMCA and Child Porn cases. This is because Deduplication is a feature of any large file sharing entity gmail included as drive space is not free.
Because of deduplication there will only ever be one copy of the relevant file clusters in existence and a table of assignments for which messages and or accounts to apply it too. Thus given an example of the file or the list of cluster hashes and a simple court order a company can expunge the one copy and/or return the list of holders with their association / upload / download dates.
Now one key issue would be that even a single bit changed in the file (mentioned in the article) would change the file hash and probably 50% of the bits in the specific cluster would flip. But for larger files >10MB it may be sufficient to match a percentage of cluster hashes and then inspect the misses further.
That said a savvy antagonist would recognise the above and suggest ways to defeat deduplication, even without using anything fancy. For a text file, simply running it through a compression algorithm would change it sufficiently and if you use one that does encryption correctly then each encipherment, even with the same key, would result in a different file. Plus since you are not actually interested in securing the file you could include the password as the filename.