I don't know. AGA is obviously less powerfull than the Cirrhus Logic chip from the Picasso2 and GVP Spectrum board.
I forgot to mention that Fusion in direct mode gives exactly the same results at benchmarks than Shapeshifter direct mode with CardTrickEVD. Perhaps is there something to speed things up directly in Picasso96? Is the direct mode still related to Picasso96 settings or is it true direct banging of the hardware?
AGA has more bandwitch than ZorroII. Cirrus chip is faster when blitting, but that won't make c2p games run faster on ZorroII. Anyway all these problems sound weird... What P96 settings are you using? Do you have all caches enabled? Do you leave enough free fast ram?
I hope you don't have enabled some weird 24bit address mode.
My A4040/25 with CV64 was 6-7 times faster than the first PPC Mac with speedometer using Fusion/Shapeshifter IIRC.