It's hard to understand why MS introduced this storage format - perhaps they hadn't yet decided to provide table compression in the next version when they did this (although one would thing SQL2008 was in a pretty advanced stage by the time SQL2005 SP2 content was defined). Using a long format + compression (store the affected tables on a compressed partition in older or less facility-rich SQLServer versions ; use SQL table compression if you have the right SQL version) is going to be far cleaner and give greater savings almost always. Of course if you have a very large very sparse matrix of decimal values, and want to store it in fully expanded form (ie as if it were not sparse) in a table it's possible that vardecimal might do as well as or maybe even better than compression, but (a) I think it's unlikely that vardecimal actually would do better and (b) I really can't imagine anyone wanting to do that.