Here a modified version useful to find the dropout voltage of each cirquit. The .MEASURE commands in the simulation find the exact values.

The simulation of circuit #6 using PNP transistors for a load connected to GND, like I built it and like you can see it on the photos.

Note: right click and chose save as if LTspice does not open when clicking on a link.

The SML-A12U8T is a red sideled from ROHM. Look here for the spice model. Models of different red LEDs should also produce very similar results.

The LT1004-1.2 is a standard part in LTspice and can for our purpose be used instead of the TLV431. If you want, you can find the spice model of the TLV431 here. The TLV431_V2.lib works directly. Here is my corresponding symbol. Just copy the files in your sym and sub directory.

To make circuit #7 work, the TLV model is unavoidable.

It should also be possible without the error log (and without the .measure Ix find... commands but unfortunately, for unknown reasons, it is not.

Well, the result of the simulation depends primarily on the accuracy of the models and a realistic circuit (which contains also parasitic components where needed).
20 cm of wiring to the load contribute roughly 400 nH of inductance, which sounds like close to zero but cannot be neglegted under all circumstances.

Just as Mike Engelhardt (the creator of LTspice) once told me, LTspice simulates the physics. So if your simulation does not show the behavior of your real circuit, you did not simulate your circuit.

Not my circuit? Well, modern models are usually relatively sophisticated and accurate (unfortunately not all models available for download are modern). But have you thought about the trace inductance? Did you put in a few tens or hundreds of pF of connection capacitance? With some circuits, you'll be surprised how that affects the simulation and how accurately the simulation approaches its measurement! I've actually had circuits that started to oscillate unintentionally and thought they certainly don't in simulation, but in fact, after a really correct schematic, they oscillated at almost the same frequency as in reality, even the trace shape was virtually identical.

In our case, this may happen due to incomplete models that e.g. do not contain the influence of temperature. At least the basic models I used work correctly. You can directly measure V_BE or the diode junction and see the -2 mV/°C. Another point may be the LED since we use it at a much lower current than in normal applications. Besides this, we do not have tricky things like saturating inductors or high di/dt, which would make it necessary to consider trace and terminal inductances.

Nevertheless, it must be made clear that the circuits here are of course idealized and assume that there are no inductive components and parasitic capacitances either in the circuit or in the supply line to the load. In practice, one circuit or another may be prone to oscillations after all, and compensation measures may be necessary, especially for those with the TLV431. However, if you model them correctly, the simulation will also tend to do so. Please refer to the data sheet for more information.

I did not verify the circuits (except #6) in real life but I suppose, they behave quite like in reality. To be sure, you would have to build them.