XML is very slow to write, and you are probably limited by that. Switch to binary mode, which is much faster. You will see in the end a maximum rate of ~500 Hz, and thus a dead time of 2ms, independent of the sampling speed. Note that you have only an evaluation board, which is optimized for ease of use. If you develop your own electronics, and do optimized readout, you can bring the deadtime down to 30ns x number of samples + 2us, or 32us if you read 1024 values from one channel.
I am currently trying to figure out how to properly characterize the dead time of the DRS4 board. My most recent experiment to try and answer this question involved using an external trigger that can range from 1Hz to 2MHz. I fed this trigger into the DRS4 and collected 1000 samples with no input to any channels. I repeated this across the range of my external trigger by a factor of ten [10Hz, 100Hz, 1kHz...etc]. After I had saved these runs in XML format, I looked at the difference between timestamps on the events. Attached are my findings. Can someone offer an explanation for the periodic peaks? I am new to the DRS4 and don't really understand how it works. My guess is that there is a buffer that has to be emptied every so often, but if so, the buffer emptying time varies with the frequency of the trigger. I would ideally like to be able to know the relation of the dead time to a particular setting I change on the DRS4 such as locking the sampling speed or changing external trigger frequency.