Errors caused by Latency or something else?

Mesa cards ( and all other servo/stepgen hardware for LinuxCNC AFAIK ) only use a servo thread and for
most applications a 1 ms (1000000 ns) servo thread is appropriate.

Typically only parallel port setups use the base thread and in this case optimizing the timing based on the
base thread latency makes sense since this will determine the maximum step rate (and step smoothness)

As you found, 37 usec is not enough time to even complete the communications with the Mesa Ethernet card.

Put like that it makes allot more sense to me.
Guess I didn't see the scale of what I was working with, and just thought of lower = better.
Again thank you, and thanks for taking the time to explain things. I really want to learn more linuxcnc (as well as linux in general) as it seems like a pretty powerful and valuable tool to be able to use.