Well, obviously a crowbar can't ride through something and self-recover, that's a pretty big disadvantage I'd say.
Crowbars were effective back in the day, because thyristors were the only available active devices with adequate ratings; in some ways, they still are today: while comparable MOSFETs are available, they may be prohibitive in cost or size. Along with diodes, they hold the impressive distinction of being the only semiconductors with a fusing rating. Enough even to short out a DC filter capacitor, mostly dissipating its contents through its ESR. So they were a common sight on power supply outputs (or internal common rails, as the case may be).
Which was especially important in the early days of SMPS, when controls were weak, if present at all: a 2-4 transistor self-oscillating circuit might be all you get. Saturable reactor / magamp designs weren't uncommon, either (and still weren't, for certain applications, all the way until very recently*).
*ATX PC PSUs need 12, 5 and 3.3V outputs, with pretty reasonable regulation on each, prioritizing 5 and especially 3.3. You can't just throw all those onto a common power transformer, rectify and filter, and have it. Instead, they get 5 and 12 from taps, regulate the 5 and 12V as a weighted sum (making their combined response quite stable, but leaving them a little squishy in cross-regulation), and for 3.3V, they take one 5V tap (making 2.5V once filtered), and half of the other, but a variable amount -- using a magamp, which is pulled into saturation using a TL431 controlling a PNP (to clamp a variable amount of below-GND flyback from the magamp choke). Thus, the 3.3V rail is 2.5V plus some pulse width of 0-2.5V, giving it adequate control range, and the TL431 means it's more accurate than anything else on the supply!
I forget if they still do that today (magamp regulation of 3.3V); most now are 2-switch forward converters -- making poorer use of the transformer (it's half wave), but the improvement in Fsw and EMI is considerable. You see same-size transformers doing 1kW today, that struggled to do 200W back then.
There's an old Unitrode appnote or two, discussing this (magamp) technique -- basically you can have an open loop (give or take current limiting/fault protection) full-wave forward converter at the primary side, transform it, and let each secondary manage how much it wants to draw from that.
Magamps by the way, act, very, very roughly, like a magnetic thyristor. In that, you basically use them for phase control; they're just programmed kind of weird. When "pulsed DC" is applied (such as from a diode from a bipolar square wave), the choke (this is nothing but a single winding on a "square loop" core material) absorbs the flux up to some point, then it saturates fairly suddenly (can be fractional µs!), and starts conducting with low impedance (essentially the inductance of the winding as if it were now air-cored). When the pulse ends, the stray inductance can generate a little flyback, but the core mostly stays in place -- it generates very little flyback, much less flux than the initial gulp in any case -- so remains magnetized, and on the next pulse, it takes very little time to saturate again: delivering full power. So it's full-on by default, and can be turned off by forcing some reset flux. If a negative bias current is applied during this reset phase, a flyback pulse can be coaxed out of it -- how much, depends on the applied current, thus making the single-winding magamp a transresistance amplifier (current in, voltage out). Or if applied voltage (through a clamp diode, to avoid drawing current from the AC supply), then the amount reset is (reset time) * (applied voltage), which will exactly match the positive output (once averaged by a filter), thus making it an inverting unity-gain voltage amplifier. By arranging feedback (in shunt, as in aiding/opposing the reset voltage; or in series, as aiding/opposing the reset current), you can get other characteristics, like useful voltage gain, or hysteresis.
(Aha, once again the diversion grows longer than the original comment
)
Tim