> I think the main "pro" of this approach is that it doesn't use any > nonstandard SQL features, so you preserve your options to move to some > other database in the future. The main "con" is that you'd be buying into > fairly significant rewriting of your application code, since just about > every query involving these columns would have to become a join.
Well, I don't really think I will move from Postgresql anytime soon. It's just the best database for me. Rewriting code is one of the things I'm doing right now but before I touch database, I want to be sure that the choices I made are good.
If your applications are read-heavy and only have a small-ish amount of code that inserts/updates the table, it may not be that much of a rewrite. You can create a integer/varchar table of key/values, use its key to replace the current varchar column, rename the original table, and create a view with the original table's name. Code that only reads the data won't know the difference. And it's a portable solution.
I did this and it worked out well. If the key/value pairs table is relatively small, the planner does an excellent job of generating efficient queries against the big table.
