Wouldn't it take the "man in the middle" quite some time to figure which encryption scheme is being employed?
To me, this could be as nasty as actually trying to crack the scheme.
True enough, but that technique, also known as "security through obscurity," has a bad reputation among cipher geeks. The "proper" way to design a crypto protocol, according to this community of geeks, is to make it "impossible" to crack, even if the bad guys know the protocol. This describes all the popular encryption protocols today.
So, in the scheme where the keystream is generated by a PRNG, this man-in-the-middle, even if he knows how the scheme works, will be required to test a huge number of possible seed values. Trial and error, but requiring a prohibitively long series of trials. Compare this with XORing always with one ASCII character, such as "a" and you can see, the series of trials will be very limited. It will probably take only two attempts, at most.
Symmetric key stream ciphers are perhaps the fastest of all encryption and decryption protocols. A variation of this is to use a block code, and then create the keystream out of the entries in each block. Another scheme is to create blocks of plaintext, but that's not as fast. Thing is, in all of these, the XOR operation is super common.
So, anything you try to devise, which ends up being repetitive, in any "security through obscurity" technique, a dedicated hacker will figure out, sooner or later.