You guys seem to have the topic well covered, but in case anyone reads this and get confused let me say explicitly what everyone else is already saying implicitly.
There's two different goals to circuit protection accomplished by different devices:
1. You want to prevent short circuits from developing large currents in the layout when possible.
This is what breakers, fuses, etc are for. Large currents stress the power devices (switching FETs in an DC-DC chopper like the one in a Z4000) and stress the transformer windings in a step-down source like a PH180 brick or other coil-based source. Prolonged short circuits tend to destroy power devices and power components but generally don't damage the electronics since in a regular derailment short circuit the large current doesn't flow through locomotive electronics. One exception is if you have power coming in through an MTH TIU (in that case the current does flow through the electronics). Generally these shorts are at low frequency (60 Hz and down) and you can just treat the wire like a wire (same voltage everywhere it's connected at the same time) which is why you can place the current protection device any old place and it's fine.
2. You want to avoid kickback voltage... high voltage transients developed in the layout from normal train operation...
Motors are inductive and inductors produce voltage depending on the change in their current. VL = L di/dt. Big change in current makes big voltage. If your train bounces along the rail making and breaking contact or it has momentary disconnects going over switches and stuff, the Ldi/dt term is big since di/dt is discontinuous going from current to no current and back to current again. Those discontinuities generate voltage transients well over 100s of volts that have duration in the 1-100ns type of timeframes. These transients are not defects... these are part of normal train operations... but you need to stop them from propagating and damaging electronics which are mostly semiconductors like the ACT244 driver in the TIUs, or GRJ's buffer circuit and like PS boards in the locomotives themselves, (all have breakdown voltages in the neighborhood of 10-20V). This is what the TVSs are for. The big difference from current protection is these short 1ns voltage transients are a broadband event, (the Fourier transform of a 1 ns pulse is a sinc(w) going all the way out to a GHz.
Since these transient voltage pulses are GHz range they have wavelengths on the order of a few inches meaning they can bounce around a layout like a transmission line and the TVS you put way over here by the station won't suppress the transient over there by the train yard since the transient voltage pulse is physically only a few inches big wherever it currently is. (wavelength = c/freq). For that reason you have to put the TVS device physically near what you're protecting (well within a wavelength).
Bonus Topic: Loading
Both DCS and Legacy/TMCC are pretty high frequency signals (broadband even in DCS's case), and while @gunrunnerjohn's booster and the TIU act244 driver do a pretty good job of driving loads, it's not going to drive fractions of an ohm. If you put too many TVS diodes everywhere or TVS diodes that are too too big, that capacitance starts to add up and eat into the voltage swing of the superimposed signalling leading to signal integrity problems through both RC loading (bad time constants sloping out your waveform) and LC loading (unhappy ringing waveform).
Also, I'm not dead. Mars helicopters and model trains are competing for my time, and sadly the trains haven't been winning.
