The reason for the 16.5 MHz is the following:

After each block of 32 bins, the DRS4 chip switches an internal segment, which causes some small spike at the analog output of the chip. This spike is a bit wider than 30ns, so if everything is digitized with 33 MHz, then you see small spiked each 32 cells. The appropriate solution would be to modify the firmware to digitize all cells with 30ns (33 MHz) and all cells with the spike with ~50 ns (20 MHz). If you do the ROI readout mode, you don't know for the first 10 cells if one of them belong to this class, since the cell address takes 10 cycles to be read out. So you would first have to read 10 cells, and then if you realize that one of them is one of the problematic ones (cell number modulo 32 is zero), you have to re-read the first 10 cells, and digitize the problematic cell with a longer settling time. Now this is a bit complicated to implement in the firmware, so I was just too lazy to do it and decided to digitize everything with 16.5 MHz. But if you are worried about the dead time, you should consider implementing the mentioned algorithm.