They quote a roundtrip of 3.6ms, so one-way 1.8ms. In the plots on their page, it looks like the processing latency is centered on around 1ms:
> The LATMON pulse width is therefore an accurate measure of the total processing latency of each cycle and is affected by every element in the data path: processing delay in the microcontroller, network transmission delay, host OS delays, signal processing delay in the DAW, etc
You are confusing the audio latency with the processing latency. I see this mistake a lot.
One way audio latency is about 3.6 ms divided by two = 1.8 ms. This type of latency is audible.
Processing latency is just the amount of time it takes to complete all processing for each audio cycle. From the histograms, the typical processing latency is about 625 us and of course there is some jitter (almost all of the jitter comes from the Linux host BTW). Processing latency is not audible. However if the processing latency exceeds the deadline on a given audio cycle, there will be an underrun which will cause an audible glitch.