I'm really enjoying using Processing. I'm having so much fun, I'm honestly waiting for the other shoe to drop... that I'll eventually start to tax the processor and memory of an average machine.

For example I've been playing with pseudo-cell shaded sprites and I'm really digging the result:

http://npoly.com/cellShade/

But in order to do something like that I have to have twice as many sprites drawn as there are gears on screen (6 images 3 foreground and 3 background).

So my question is... are there any things to bear in mind, performance wise, when building bitmaps for use in Processing?

I imagine keeping graphics as powers of 2?

Anything as it pertains to png vs jpg (assuming both are similar filesize)?

I like to use alpha transparancy, is it a significant resource hog compared to binary transparency in Java/Processing?

Any other tips?

Once loaded it doesn't really matter what format bitmap file is in because they are all stored as PImage objects in Processing. PImage stores the graphic as an array where each element represents a pixel and each element is a 32 bit integer representing ARGB values.

If the image is png of gif then it will recognise any transparency used in the image and apply it, obviously jpg files does not support transparency.

You can manipulate the actual values in the array to get a number of nice effects but depending on the image size this canbe quite CPU intensive.

Not sure about the powers of 2 although this is normal when used as textures in 3D stuff.

You could look at
http://processing.org/reference/pixels.html
and
http://processing.org/reference/PImage.html
for more info.