PAW is able, before starting the loop over the events, to find out which are the columns of the CWN which are actually referenced in any given query (command line selection expression or COMIS routine). Only the columns which are referenced will be read from the file.
If you analyse Ntuple data without PAW, it is your own responsibility to find out and specify which are the columns referenced via the routines HGNTV or HGNTB (if all variables in a block are needed).
This optimization can be very important. Assume, for instance, that a 100 Mbyte file contains an Ntuple consisting of 300 columns, and that 3 columns are referenced. Then without optimization the complete 100 Mbyte file will have to be read, while, by specifying the columns used, only 1 Mbyte will be input.