In a variety of use-cases, referring to historical files by name can be ambiguous. Either at the time you're referring to it it had a name other than what you're putting in, or a new file currently occupies the same name, etc. While bzr's behavior may be well-defined, it can be difficult to work out what to enter to get the desired result.
It would be nice for historical-looking commands that deal with files (log, diff, stat, cat, probably others) to be able to accept a file-id to unambiguously refer to a given file. How to do so is somewhat problematic.
One option would be to try to look up un-findable file names as if they were file-id's, allowing file-id's to just be entered on the command line like file names. This has potential collision issues, but only if you have a file named like the file-id you're entering (which is unlikely, though not impossible). Other possibilities exist.
I would probably prefer to have a flag indicating you are passing file ids. Also, we probably need to make it easier to get file-ids if we are going to start taking them as parameters.