The best take-away is this:
If you wish to have a PSU that's as gentle as possible, it must have a modest impedance at AC frequencies (so that it is neither a very good voltage or current source), and a controlled, dependent impedance at DC (so that, after some time, say microseconds to milliseconds, it "makes up its mind" and stabilizes to a constant setting -- voltage or current, as the case may be). With minimal overshoot when crossing between the two conditions -- a part of the "soft AC" requirement.
Indeed, putting a capacitor at the output of the power supply might be exactly the wrong thing to do, for many reasons.
The best reason not to is because it's simply futile! ...Or should be.
Suppose you're testing with some average banana jack cables strewn about the bench. Each is 3' (~1m) long, so you've got very roughly several uH of stray inductance spread about (depends on how "strewn" they are..). You can never possibly hope to have a well-bypassed and stable voltage at the ends of those cables, no matter how much capacitance you put at the beginning!
What's worse is: if there's a bypass cap at the PSU end of the cables, it presents a very low impedance, making an effective AC short. So both the +/- cables act in series as one solid inductor, and plugging that into a circuit that has just a few bypass caps (a few 0.1's, say) will ring like one hell of a bell when suddenly connected! If you set the power supply to +/-15V and connect it to an analog circuit, don't be surprised if it explodes from the +/-30V peak transient!
Overshoot is lessened if the circuit has electrolytic capacitors, because their ESR tends to be dominant, and comparable to the ESR and reactance of the capacitor inside the PSU too (if applicable). But this brings a second problem: if you have anything electronic in-line with those power cables, the inrush current can easily be in the >100A range, and pop goes -- diodes, transistors, whatever. Shorting capacitors together is an extremely stressful event, and if a semiconductor has to bear that transient, it won't much appreciate it!
So the whole point of putting that cap there is to make it a "nice stiff" supply. Well considering you'll never have a circuit plugged directly into the binding posts with zero lead length, that's a useless concern! And anyway, a properly designed power supply should be as fast as an electrolytic capacitor, so it should be redundant in the first place!
For reference, I have a bench supply, which is more of a high current audio amplifier than a power supply as such. It's got +/-25V rails (with well over 10A DC available), complementary darlington outputs, and over 10kHz bandwidth. It would make a fine subwoofer amplifier, if you have 1 ohm speakers handy. I normally use it as a power supply by wiring a variable DC reference to the input. I normally leave this beast on the floor, and connect to it through a ~6ft (1.8m) twisted pair cable. I measured its short-circuit transient output at over 40A. It took some 10 microseconds to climb up to that current, largely because of the sheer inductance of the cable -- about 2uH. Since V = L * dI/dt, for an applied 20V, it should take 4us to reach 40A -- the additional time being due to resistance in the circuit, and the actual response time of the amplifier. The delay is longer with more cable, and less without.
This is the kind of performance you should be shooting for -- any faster and it doesn't matter, because no one will be using little enough cable to matter. Any slower, and you will see the potential consequences of overshoot, or short circuit transients, or stuff like that. This can be more trouble in some designs than others -- a switching supply might be limited to run much slower than this, for a chain of reasons; and probably can't avoid having some amount of capacitance (often a large amount!) on its output, for filtering. In that case, you're kind of stuck. A hybrid (switcher + linear postreg) might be okay, but might also have some unsightly side-effects (like slow recovery following a short, as the switcher recovers), and the challenge is to hide those as gracefully as possible.
Tim