DPC = Deferred Procedure Call
DPCs are typically setup by interrupt service routines (ISR) so that they can be executed outside of the context of an interrupt.
DPC Latency is a measure of how long the OS takes to execute a DPC once it has been created in the DPC queue.
For example, a sound card will typically buffer 128, 256, 512, or 1024 audio samples per channel at a time. 1024 samples at 48,000 samples per second is enough for about 50 milliseconds of playback. As this buffer runs down, it needs to be topped by the sound card's software driver. The sound card will interrupt the CPU, and the CPU will execute an ISR in response to that interrupt. In order to keep the ISR as short and sweet as possible, all that the ISR does is create a DPC to refill the sound card's buffer. At that point, the ISR is done and the CPU can go back to its regularly scheduled programming. The OS will execute the DPC at a priority level that is high, but not high enough to clobber more important interrupts. If the sound card's sample buffer depletes before the DPC executes and refills it, then the sound card will have nothing to playback. The result will be choppy audio.
A 1,000 microsecond DPC latency is a bit high but the important part is that it's flat. It's not all over the place. Periodic spikes in the DPC latency indicate that there are poorly written drivers with ISRs that are taking way too long to execute.