Why should we expect user-space coroutines to be lighter than OS threads?

I have heard it said that managing concurrent TCP connections using user-space coroutines uses less resources per open connection than using an OS thread per connection.

Actually, OS threads and coroutines are essentially different ways of representing the same thing: a combination of control and state.

An OS thread stores:

• A pc counter (representing control)


• A stack pointer + other registers (representing state)

A coroutine requires:

• A function pointer (representing control)


• An environment pointer (representing state)

Why should we believe user-space coroutines to be a more resource-efficient approach?