The diode is in the best position, and is of an appropriate type.
It conducts when the input is negative, the same as the transistor base conducting when the input is positive. The 47K resistor is about 1/10 of a normal RS-232 load. One could also block the voltage, but then a -100V spike (ESD say) could break down the 1N4148 and break down the E-B junction, causing irreversible damage.
Also, a 1N4148 is an appropriate diode for this application. It's a "switching diode", low capacitance and fast reverse recovery. A 1N4001 would also likely work okay, at least at slow baud rates. The 200mA rating means that even if a very high voltage were to appear at the input the transistor is fully protected, at least up until the resistor arcs over,.
Forget about USB for a moment, because these devices are useful in a lot more applications than just USB. First, the diodes in your first two examples would not usually be generic signal or zener diodes. They are usually purpose designed transient voltage suppressor(TVS) diodes.
TVS diodes come in all different flavors, but there are three key parameters. First, the maximum reverse standoff voltage. This is the working voltage. For USB, you'll want to be very close to 5V. The next thing you'll want to look at is the clamping voltage. Again, around 5V for USB. Usually, the leakage current rises significantly as the clamping voltage approaches the reverse standoff voltage. Be aware of that in case it matters in your application. The last key parameter is the peak reverse current. Make sure it can handle the standard you're going to test against.
There is one more thing to be mindful of when placing TVS diodes on high speed signal lines (ie. USB, HDMI, Ethernet, etc). Make sure the TVS diode is low capacitance. There a many diodes and arrays suited for this purpose and will have capacitances in the neighborhood of 0.2pF.
I'm going to skip straight to your second example since we're talking about USB here. The scheme shown would be appropriate if the diodes shown clamp around 5V, and are trying TVS diodes. Regular zener diodes are not fast enough for ESD protection.
In your last example, you question whether the top node is connected to Vbus and according to the device datasheet, it is not. If a positive ESD event occurred on any of the signal lines, that TVS would clamp it. The Vbus TVS has a higher clamping voltage for the power rail.
Best Answer
I think this is a difficult challenge to get decisively below 0dB and this is what i found along the way: -
Here is a decent article from SkyWorks about protection using PIN diodes and for lower frequencies using schottky diodes. It's pretty much encapsulated in the diagrams below:-
If you look at the 2nd graph down (single stage limiter using PIN) the maximum level coming through is going to be about 11dBm and this is 10dBm above your receiver limit.
However, the 2-stage detector limiter is going to operate at about 0dBm so maybe this is what you should be implementing. Read the blurb!
If you look at the Avago ASML-5822 it is a single stage limiter with schottky and it can only manage about 3dBm limiting so it looks like a 2-stage circuit is your best bet unless you are operating sub UHF where you will probably make a limiter with back-to-back schottky diodes as per figure 4 on the picture. However, it's going to be about 1.4Vp-p which is still about 10dBm so good luck hunting.
I'm at the extent of my knowledge on this area so I'd wait for an answer that makes better inroads into getting the limiting power sub 0dBm. I'll be doing the same.