Back when the XPG3 came out and we were trying see if the dedomed or sliced version had good performance, I was learning about the structure/architecture of LEDs better in order to understand the connection between dedoming and light recycling. See here for an explanation.
In this article , some details of different architectures are discussed. It mentions one strategy for increasing light extraction efficiency, light recycling, in which there is a reflecting silver layer below the LED junction. Photons that don’t escape the LED package are reflected down past the junction and reflect off of the silver layer, getting another chance at exiting the package. I think the luminance increase we see from dedoming has to do with this reflecting layer.
As we have noticed, there are some LEDs that don’t have bond wires on top. This is an architecture called “flip-chip”. Both the positive and negative contacts are on the bottom. My hypothesis is that the flip-chip style LEDs do not benefit nearly as much from dedoming as the traditional architecture with visible bond wires because of a difference or absence of the reflecting layer under the junction.
For example, the XP-G3, which is a flip chip style, did not have good performance when sliced or dedomed. The Nichia 219C which also is a flip chip style without visible bond wires also did not have good performance when sliced. The XM-L2, XP-L, and XP-G2 experience a ~1.9x increase in luminance upon dedoming, while the flip-chip style experience a ~1.3-1.5x increase. Another example is the third generation Oslon Square, which has visible bond wires. Although this LED is apparently difficult to dedome, user FmC has had some success, and the luminance increase is on par with the XPG2.