I don't think I have ever read so much incorrect information in one place before. Congratulations! By the way, I have over thirty years experience doing hardware design and assembly programming.
First, to your addressing question. I don't know if you are talking about segment-register type addressing, or bank-select type addressing, but in either case you are completely wrong. In degment register addressing, the processor performs the calculation of merging your 16-bit 'address' with the current segment, and drives the addressing lines accordingly. In bank-select addressing you pre-select a 'bank' of memory, and the addressing lines from the processor select the appropriate location within the bank. In either case, if you can address 4G then you have 32 addressing lines, either all directly from the processor, or perhaps with some coming from an external bank-select register.
I already said it does not matter what the unit is being selected (bit, byte, word, line, whatever). The addressing does not change based on the size of the data, on the number of data lines changes.
WTF does something being measured in bits-per-second have to do with powers of two? Not a damn thing. Bits can REPRESENT powers of two, but they do not OCCUR in powers of two. If you can't understand that distinction you are really even more clueless than I thought. In memory components, bits/bytes/whatever OCCUR in powers of two. You can't buy a 1000 (not a power of two) byte memory chip, you can buy a 1024 byte (a power of two) chip. You can't buy a 3072 byte (not a power of two) chip (well maybe there is some weird chip like that but it would be special purpose), the next highest size is 4096 (a power of two). However, you certainly CAN send exactly 1000 or 3072, or any other number of bytes across a network. There is absolutely no power-of-two boundary involved there.
Likewise, the size of a harddisk is dependant only on the bit density of the medium. A disk can be manufactured in absolutely any size at all, there is NO 'natural' power of two boundary to disk sizes.
Lastly, grouping bits into 'bytes' or 'words' has nothing to do with powers of two, it has to do with MULTIPLES of bits. There is no reason that the 'word size' of a machine has to be a power of two. IBM mainframes use 24 (not a power of two) and 31 (not a power of two) bit addresses. There have been 6, 10, 12, and 18 bit 'words' in the past. None of those are powers of two. The only thing grouping into bytes does is say that you will always transfer or store a multiple of 8 bits, which has nothing at all to do with powers of two.