Recently I've been doing performance tests on the iPad and on 3rd gen devices, and to my surprise it seems that:
a)
A8 textures are (8-bit grayscale textures) are MUCH slower than 16-bit and even 32-bit textures.
A8 textures should perform faster than 16-bit textures, and this is true for 1st and 2nd gen devices, and I don't know why it performs MUCH slower on the new devices.
b)
CCPointParticleSystem is MUCH slower than CCQuadParticleSystem.
Again, point particle system occupies less memory and performs less calculations, so in theory it should perform faster than the quad counterpart. On 1st and 2nd gen iPhone devices it performs a little bit faster.
My guess is that is has to do with the OpenGL ES 1.x emulation that is included on the iPad/3rd gen iPhone devices.
Updated Best Practices: http://www.cocos2d-iphone.org/wiki/doku.php/prog_guide:best_practices
See the v0.99.3 performance test with the iPad results: http://www.cocos2d-iphone.org/wiki/doku.php/perf_test:0_99_3
So, either you can check the device and runtime or, the "lazy" approach would be to always use a Quad particle system.