Adreno SOC have bottleneck, the bandwidth between CPU and Memory suffer of slowdown. is it software or just the mmu which is not as fast as SGX or Tegra processor i don't know. Now why Big 3D games do not suffer while PS emu do, it's because we have no choice than transfert many textures a second, while Games store their textures into memory and will just transfert new textures when scene change.
ON PS1 emu it's different, far from OpenGL architecture, the trick on PSemu is to apply many thing and store a lot of texture which just represent for example the same object but with different palettes. but that's not all, just take sprite, if somes effets are applied on them, it's more textures stored. that's take Many space in memory and bandwitdh must be high. MMU is very important too and with fullhd screen, framebuffer take 4x more space than 720p. OpenGL plugin use a garbage collector to remove dynamically textures used frames ago to store new one, but this operation is managed differently regading GPU... on MALI it's not handled immediatly but when memory used reach a limit.
With fullHD screen, MALIT450 would be good, SGX544 is surely good enought too, while Tegra 4 run very well too.
Anyway i know a good method to boost this, i need to experiment it.
Just pushed FPse 0.11.118 to fix the immediate crash on Cyanogenmod kitkat.