The text is written as if it's thinking about handling 64-bit
unsigned long, but it doesn't.  The tables that map bits to
pixels are completely unprepared for this.

I thought about widening the tables, but I thought they'd get
unreasonably large.  There's no reason not to go ahead and 
handle this 32-bits at a time.

I thought the tables would come back to haunt us. The only reason the 
tables where introduced was for speed enhancements. I reason the code is 
extra complex was so you could pass 64 bits of data across the bus. I 
still like to see that happen still. What do you think?