If you suspected a file might be intercepted, an archive with a password would at least make the contents difficult to recover, but certainly not impossible. It occurred to me that most methods used to crack a passworded archive would quit early if a password worked, so one or more dummy weak passwords could be used to obscure the real content. Of course obscurity shouldn't be the first line of defense, but it can't hurt. Can this be done?
in theory it can, in practice I doubt it. In essence, you would need to craft a message that would have a double hash-collision, one for each password.
Suppose you can craft a meaningful msg that would decrypt using password P1 into PlainText1 and using strong P2 into PT2.
The problem - crafting PT1 and PT2 such that P1(PT1) == P2(PT2) == E.
Edit: @Banaguin raises some good points - if P1 == PT1 and P2 == PT2 you can craft such an equality. The problem is that P1 should be a "weak/trivial" password, while P2 should be a "strong" password.
Using P1 == PT1 and P2 == PT2 would not scale for large messages - a 1k text message would become non-trivial to decrypt even for the "weak" password and with lesser probability it will be chosen over the "strong" password.
The second problem with this approach is that the "strong" password should be "strong" - if it is the message itself, there is no guarantee that it will be strong, unlike properly crafted passwords.
What you describe is not possible with the common archivers. However, steganography was invented pretty much for that purpose.