As laptop2d and transistor say, NOISE! That's the key concept. Everything else is just signal (sorry, I couldn't resist that).
Back up a minute and simplify the question. Consider just an unchanging voltage on the wire, supplied at one point and measured at another. And let's say that the transmitting (supply) end is a theoretical textbook-perfect power supply. The transmitter drives a certain voltage on the wire. At the receiving (measuring) end, we ask the question, "What is the voltage?" If all we can say is, "Eh, about 1.2V.", that's not a lot of information. If we can truthfully say, "It is 1.23798570520664V, give or take a few femtovolts", then that is a lot more information. In fact, if there was no limit to the precision with which we could supply and measure a voltage (and nothing interfering with it along the way), there would be no limit to how much information we could stuff down a wire.
In the real world, our precision is limited because of noise. There is noise in any real voltage source, noise due to thermal excitation of the wire, noise in the measuring instrument, and so on. There are also other practical limitations because of real-world tolerances on component values, reference sources, and more, but we don't even need to think about those to see that we can't supply and measure a voltage with unlimited accuracy.
So, how do we get more bits down a wire? We trade off accuracy and time. For example, instead of trying to measure a voltage once at femtovolt accuracy, we split the data up and make multiple measurements at, say, millivolt accuracy, spread apart over some period in time. Now things start to get really interesting, because there are so many different ways you can do this, and the reason there are so many different schemes in use is because they each make a different set of tradeoffs with regard to speed, complexity, cost, power, robustness, and what have you. Also, as others have mentioned, once you throw the time element into it, you have a new set of problems in addition to pure noise, because there are distortions and reflections that limit how quickly you can make changes at the source and get a distinct result at the receiver.
So, to get back to the original question, "How much binary data is transmitted across a physical wire in one Hertz cycle?", it all depends on how you transmit and receive it. If you want to dig deeper, the field is called "Communications Theory", and Claude Shannon is its patron saint.