This is an excellent question that I have been asked a number of times lately since I work for an address verification company called SmartyStreets.
First off, a postal address represents a single locatable point on the map. An address by itself is inherently benign because it doesn't have any additional information. Drawing a point on a map doesn't do anything. It is only when you begin to assign CONTEXT to that point (address) that it starts to mean something.
With that in mind, a postal address can represent a person, an organization, a building, a car, whatever. Once you start gathering multiple postal addresses you increase the context that can be derived from that grouping. Similarities can be determined to see what the addresses have in common. Still, just a grouping of addresses in a like area doesn't denote much context. I can look at a google map and see all the houses in a certain area. That's not a breach of privacy unless I have unauthorized access to privileged information.
Other points of context must be combined in order to actually give away any kind of private data. For example, a group of postal addresses that are submitted to an online service for address verification and/or geocoding doesn't give away information unless you know who submitted the list for processing. Once the list owner is known certain inferences can be made about the intended use of the list. Knowing this additional context, such as list owner and intended use, would certainly qualify as privileged information and can be a source of privacy breach.
Bringing the processing "in-house" so no external data service is involved is an option. It certainly excludes any type of unauthorized access to privileged information. Address verification and geocoding are not tasks for the uninitiated and certainly require advanced skills (meaning experience gained over time) in order to process very large lists without consuming inordinate amounts of time and resources. So bringing it in house is certainly an option, but does every company that has sensitive address information have the resources to do their own "secure" address processing (including geocoding) in house? No. (Although it would certainly mean job security for the readers of this website.)
There are ways to maintain the requisite privacy and still use online services. One method would be to create an account, get everything tested and figured out and then, using a temporary email address, set up a new account with an unrelated billing address associated with a credit card that can't be traced back to you. Processing the addresses on this account would theoretically not give away any valuable context and thus would maintain the privacy of the individuals on the list. (This is starting to sound like the movie Enemy Of The State.
If that sounds complex and unnecessary, I agree. A simpler method would be to take advantage of an API that uses HTTPS and POST and that doesn't store or log any of the data that you process. The use of HTTPS means that the only record would be a timestamp and the IP address that you call from. The underlying URL would not be known. Of course the account that you use would lead back to you BUT, that's not a problem because using a POST request allows you to attach a payload (in this case a batch of addresses) and the contents of the payload are not logged. Thus, the addresses that you submit are not on any server log. And the fact that they memory is wiped between each process means that those addresses aren't ever stored or logged and their transmission back to you is done over a secure connection. The end result is a log like this:
13Mar2012 06:31(-6) IP:220.127.116.11 UserID: 875564 -- POST QTY:3439942 -- [Processed]