In my experience, using better matched LED's e.g. +/- 0.1V in parallel will not result in thermal runaway, whereas std tolerances +/- 0.3 or more may cause failures.
But when a series string of identical LEDs with the same std tolerances tend to have lower deviation within a batch, you can safely parallel series strings without having to add bulk resistors to dominate the internal bulk resistance in the LED. My experience showed me that the number N LEDs in series needed to avoid adding series R has reduced in time as LED process controls have improved the quality over the last 4 decades. At one time N=1 meaning every single LED needed at least some external Rs to avoid current hogging or current sharing and thermal runaway issues. Today it is possible to have tolerances within a batch low enough to do this without adding Rs if you know what the tolerances are. I think there is a TI technical bulletin which supports this characteristic. Also for thermal assistance,
https://www.wikiwand.com/en/Thermal_management_of_high-power_LEDs
This critical thermal runaway threshold margin drops as you increase the current to maximum so you need to understand the thermal resistance- tempco -4mV/'C and power dissipation to determine when you need to add Rs to a series string for shunting in parallel.
Typically I look at the rated power dissipation of the device (LED or Diode or Vce(sat))*Ic or case package and use Rs=k/Pmax for k= 0.25 to 1 then add a bulk Rs in series that is >= to this value in series as the minimum to prevent thermal runaway at rated current and slightly more to share current equally within low %. This is why you see ganged audio drivers with 0.1 ohm more or less on emitters of high power audiuo amps using ganged transistors (N=1) .
A power 3V nom white LED will have an internal Rs of 0.25 to 1/ Pd max resistance.
eg. 1W LED 1 ohm max, 10W LED 0.1 Ohm max @ 10W as the incremental resistance.
Thus to make an array of say 100W from 12.5V using 10W chips you need to know the junction rise of each chip. This is why traffic street lights in early designs often failed with one or more shorted LEDs. The designer failed to recognize LEDs near the centre were hotter and thus lower Vf and would start to hog more current and fail in parallel.
How many street lights have you seen with burnout chips in RYG stop lights? they do not all fail from thermal runaway, but it is a common root cause.
2W White LED Ri = 0.5 Ohm, some better quality ones are 0.25 ohm.