Thanks for the link - interesting read. I looked at the specs for those board again a little closer. Looks like both the Pi 3B and C2 have a 64bit ARMv8 quad core ARM Cortex A53 CPU with 32KB L1 instruction and 32KB data cache for each core and a 512KB L2 shared cache, so no difference there apart from the ~66% clock speed boost for the Odroid, of course. However, there is a big difference in the GPUs of each board:
RPI3B: dual core VideoCore IV GPU @ 400 MHz
Odroid-C2: five core ARM Mali-450 MP3 GPU @ 750 MHz and AVE video engine
So if OpenNI is designed to take advantage of the extra GPU power in the C2 then it makes sense that's where the performance boost comes from. I don't know enough to know if that's the case though.
Both boards are 64 bit, so maybe the same type of performance increase could be seen by using the 64-bit version of OpenNI on the Raspberry Pi 3B? Or is that not possible or difficult due to the RPi's being somewhat closed source? The Driver API section of the Raspberry Pi Wikipedia page makes it sound very complicated.
Do you (or anyone else) have any experience with the Persee? If we knew what kind of framerates the Persee got with its 32-bit CPU running at 1.8GHz and Mali-T760 MP4 GPU we could compare it to the 30fps @ 640x480 you got with your 64-bit Odroid-C2 running at 2GHz with Mali-T450 MP3 GPU.