QVD is storing unique data with pointer to the actual values. Every field has its own table and pointers to that value. Pointers are bit stuffed and they are used to lookup actual values. Result of this kind of keeping is better compression. This compression is done every time when load script is running.
Does anyone know why when loading data in loops with concatenation without where or transformation is not optimized?
If you use a multi-tier architecture, you can enjoy the speed of optimized load, since you have made all of the aggregation previously and a store into qvd command, and after when you would use this qvd-s you wouldn't have to use a command which would break the opt load.