All +DSP and routing is programmed on a SHARC DSP chip.
The routing from the master (routing) DSP to secondary (graph) DSP on legacy boxes is the source of the constant 16 samples of latency through the +DSP processing. The 16 samples is only for the legacy +DSP graph and it is from the source side to the process bus side
The 2d Card actually reduces the total latency of the hardware by about 32 samples (48 samples if you were running through +DSP).
In 2D there is an additional two sample buffer (one to buffer the input from the ADC/AES/ADAT/Firewire and one to buffer the processing before it is sent to the DAC/AES/ADAT/FireWire) — so that the DMA is not using the same side of each double buffer as the processing loop is.
So it is not zero, but it is really close (approximately 40-50 uSec at 1x rates and 20-30 uSec at 2x rates).
And it doesn’t increase no matter how much processing you throw on, except for:
The delays add latency.
The Pitch Shifter adds latency.
The Reverb (non-direct sound) adds latency.
on ULN8 the latency from A/D -> Mixer -> D/A is ~67 samples. Adding character or EQ will not increase the latency