What they're trying to tell you is that the address lines in a 16 bit (or greater) system get moved up 1 bit due to word accesses (1 word = 2 bytes thus two chips) therefore the 'magic' addresses you need to access in order to erase the flash need to be altered accordingly. Thus in an 8 bit system the magic address may be 0xaa55, in a 16 bit system the magic would be this number times 2 (or shifted left 1 bit) which would be 0x154aa. In a 32 bit system (assuming a 32 bit data bus) you would have 4 bytes (assuming 4 flash chips in the system) so therefore the magic addresses would be times 4.
Also there is the issue of odd/even chips in a 16 bit system - there's two devices to erase so going on my above example the even chip would be 0x154aa and the odd chip would be 0x154ab (+1). For a 32 bit system the offsets would be +0,+1,+2,+3 for each of the four chips. Get the picture?