The point I was trying to make was that the article appeared to start from the point of view that what was needed was just an efficient 32-bit representation of a particular set of varchar(15) strings. Well, that's ignoring the fact that those strings are actually a human interface representation of 32 bit IPv4 adresses, so that the natural - and ultimately the most efficient (since real software using IP addresses often wants the real 32-bit IPv4 address, not some different 32-bit value) - representation is to use the IP address to represent itself. It was as if the author knew that IP addresses were in theory 32-bit values but were not aware that that is what they are in practise, not just in theory, that the 32-bit value is what almost all software, other than user interface presentation software, actually uses, and therefor thought that any old 32-bit pattern would be just as good a representation as the 32-bit IP address itself as long as there was a one-to-one correspondance.

With the article's representation, there is a serious possibility of dangerous misunderstandings: anyone experienced in data communications who sees a 32 bit int column used to represent IPv4 address will naturally assume that the integer is the one represented by the same bit pattern as the IP address itself; so that approach would need caveats all over the documentation and in code comments and even then would risk people taking the integer as the IP address, even though it clearly isn't. So there's an "is this way of doing it really safe" question as well as efficiency concerns.