I've been intending to take a look at this issue at some point, but am not sure when I'd get to it.

I took a quick look. It does seems to me that it is true that for data-validity purposes the message files need to be opened in binary and fed to the email package in binary.  But this is so that the message will get decoded using the correct character sets, not to avoid the decoding.  In Python3 it makes no sense to manipulate the subjects as binary strings, so the example of "looping through the messages and printing the subjects" is still going to require decoding.

There may still be ways to make it more efficient for common use cases, but that will require more detailed analysis.