I compute the picture as an array of blocks, then compute each pixel
in a block. If blocks of the same colour surround a block, then I
ignore that block. It saves some time when there are large areas of
the same colour. The picture below shows a partially completed area:

You can compute every pixel if you want to. Takes longer but you get perfection.
Or go faster tricks? Not yet in this release. The old DOS version had PC assembly code fast arithmetic of some 30 fractals, but now we have GHz Pentiums so it's no longer necessary. (The first fractal program I wrote took 10 hours to do a picture on an 8086 with no maths coprocessor!)
What else should I tell you? Right: