You know, not only does numeric "closeness" not have any direct relation on geographical closeness, with the widespread use of NAT there's not even any guarantee that requests from the SAME IP have any geographic relation to each other, in terms of the geographic location of the end user or machine making the request. There are plenty of corporate networks that are spread across the US that have only a single internet drain, for instance...
UPDATED: I thought of another couple examples using some parts of network design (meaning, not directly related to an IP assigned to Joe Blow's home connection).
First, it's common for a network designer to have a larger subnet (like, say, a /24) set aside solely to be used for point to point connections (that is, carved into a bunch of /30s). So, 184.108.40.206 might be in Chicago, 220.127.116.11 might be the other end of the link in Los Angeles, etc. Not only are adjacent IPs in often in different physical locations by design, the entire supernet might be filled with these /30s that are all in different locations.
Another similar example is loopback IPs. Again you might have a /24, but instead of being carved up into /30s it's carved into /32s (single IPs), each of which is used on a loopback interface on a router. So, similar to above, 18.104.22.168 might be lo0 on a router in Chicago and 22.214.171.124 might be lo0 on a router in Los Angeles, and the two routers don't even have to have a direct connection between them any more.