#1: Aside from a pointless top fill that's not connected to anything, this seems solid enough. (Haven't seen the bottom though.)
#2: Looks like mindless use of flood fill. Should still work alright, but...
#3: Not very good use of copper (most of it is etched away) but seems solid otherwise. Note sure about the use of what seems to be a multiturn pot, most of these tend not to be designed to carry substantial current. Mind you, you can always put in a fixed resistor of the appropriate value for your desired output voltage instead, so no biggie.
This is the dodgiest one of the bunch, I'd say. Would you trust a PCB designer who includes an immediately obvious ground loop?
I don't see any advantage in the pop-amps versions. Simple solutions are always better.
In other news, all generalizations are wrong.
What you want is enough complexity to get the job done, but no more. More complexity means better performance if done right, but also more things to go wrong. Blatant overengineering where it is not required just makes things less reliable.
Now in this case, most headphone amps (like a JLH or some opamp-based job) tend to have a more than decent amount of PSRR. A tricked-out LM317/337 combo like #3 easily provides noise low enough for something like this. (About 30 µV for the LM317 if memory serves. Realistically speaking, this means that you need maybe 40 dB of inherent PSRR, which is not a major achievement.) If you get 50 dB out of the amp and 70 dB out of the reg, 1Vpp of ripple turns into 1 µVpp, and you'd think that with enough space for up to 3300-4700 µF / 25 V filter caps, you'd think that primary-side ripple should be kept at bay decently to begin with. (The 50 dB figure would be for something like a JLH, more modern concepts may reach 80-100 dB around 100 Hz.)
One of the more advanced concepts would be something to consider for an MC phono preamp or similar.
