16 bits of one plane only equals 16 pixels in 1 bit per pixel mode. If you have more than 1 bit per pixel, you're handling parts of pixels. Part of pixel != pixel :p
Every word transferred in 256 colour planar mode is 16 * 1/8's of a pixel, in chunky it's 2 * 8/8's of a pixel. The maths works out the same, the fewer the bitplanes the more planar wins in terms of speed.
It's absolutely NOT insignificant. When you're blitting 16 pixel wide bobs (quite common), you're wasting half the blitter time on the 16 extra bits per bob.
Sure, but the overall time of those small blits is then not the blitter bandwidth but the CPU time driving the blitter. The bandwidth itself becomes more important on larger bobs, when the cpu and planar overhead decreases.
Anyway, assuming AGA chunky, I was wrong, because the blitter would still have to access whole 32bit words in memory to get to the individual bytes. That it would get rid of the need for extra shift bits is irrelevant. Only a faster blitter would make sense.
Yes the blitter could be exactly the same if AA was chunky as you still need to shift by 0/8/16/24 pixels, or you would only be able to move bobs on 4 pixel boundaries.
A 32 bit blitter would have been the only practical way to make it faster and the time and money to do that would be difficult to justify, when everyone was saying AAA was the future. Although I'd have probably left the 2d blitter to work with 16 bits for compatibility and added texture mapping, which coupled with chunky pixels would have it easy to write doom clones for the amiga. The 3do was started around the same time as AA was in development, so the idea of a 3d games machine wasn't particularly exotic.
What is the limitation of a bit-plane system?
It seems to be a flexible system.
It looks as one could design a chip to move almost any number of bits at a time.
Yes you could do 32 bit true colour displays with bitplanes, however it wouldn't be worth it.
The main limitation is manipulating images with a cpu is more complex. In 256 colour chunky pixel mode you can use a single byte write to set one colour. In planar mode you need to read 8 bytes, mask the pixel out of each and or in the individual bits of the pixel and write it back. Once you get to 256 colours it makes far more sense to use chunky pixels. The blitter overhead is also a factor, but nowhere near as much as the cpu.