Sometimes, if the enemy is trying to be sneaky, it's better to overhear them and thwart their efforts.
Other times the enemy might be trying to coordinate a massive attack for which timing is key. It wouldn't matter if you overhear because you need to disrupt the attack before it starts. In that case it's better to not let the enemy communicate at all.
And computational machinery to implement it, at least for telegraph seems to have quite likely existed in 1915: https://patents.google.com/patent/US1310719 (filing date 1917).
My only point was that by the modern standards we take for granted, OTP/XOR is not really a cryptosystem but better thought of as a primitive. IMO for those just learning crypto, it's a red herring that hides the core functionality of modern cryptography (see: the common amateur reinvention of using a PRNG as a OTP).
My answer to your main question is because encrypting layer 1 adds negligible security properties to the whole system (can't tap that link, but you can tap anywhere else), everything should already be encrypted at a higher level (for the previous reason, as well as integrity/authorization/etc), and that the bandwidth is too high (the entire point of packet switched networks is to do as little as possible in each node).
Applied to the existing Internet, the only thing it would gain is hiding addressing metadata from an attacker who directly taps a link.
A more effective general way of doing this would be something like onion source routing, where each router only knows the next hop. But once again the scalability problem, so the desire is better applied to an overlay network rather than convincing backbone providers to take this on - there is barely the impetus for IPv6.
Right, but then the Germans know you’re listening and will be more careful about what they say. If you tap the cable, you may be able to keep it a secret.