Dimensions Slightly Off

This is a fairly drastic solution but using a 48 volt power supply will more than double torque (but you would need to replace the drivers with a DM542 or similar). That will certainly fix any missing steps.

I don't know why these Chinese machines only use 24v. We are in the process of retrofitting a 2400mm x 1200mm plasma table with rack and pinion drives and a 24 volt power supply.  We are replacing the motors of unknown origin with quality ones, adding 5:1 planetary gearboxes and using Lam Technologies Stepper drivers with a 60 volt AC power supply. We use a Mesa 7i76e. We expect rapids to increase from 5 m/min to 25 m/min or higher.

I believe both computers were bought and LinuxCNC installed at the same time, the ASRock was not used until we tried to set up the router. The mill has been running LinuxCNC for a long time.

But why would 24V work just fine with the old controller? What difference does the new controller make to where we are all of the sudden skipping steps?

If the same drives are used, and the same motors, and if everything worked before, forget the power supply for now.
In the ini file, set acceleration to 100 for all axis, set the max velocity to 10 for all axis, do some tests.
Any improvement?

Check also steplen, stepspace, dirhold and dirsetup values.
Some of those may be too short and then you lose steps

Update:

Slowing it down worked pretty ok, it was still skipping enough to make a fingernail-width ledge on the skipping axis. I thought that maybe I was getting some interference, and turns out I had forgotten to connect some shield wires to ground. I connected them as close as I could to the signal wires, and that helped immensely. I can still feel a skip every once in a while, and on longer (>15 cm) moves it still has a slight ledge or "lip" from where it is not travelling to where it should on each layer - but it seems to be good enough. If there's any ideas to get rid of that skipping once and for all, please let me know. Note that it does not seem to skip when it's not running gcode. I can jog at the same speed with the spindle on and putting ~15 lbs of force against it and it travels right to the end of each axis and back perfectly.

For steplen, stepspace, dirhold and dirsetup values, I upped those to 30000 each just to be safe with not a lot of improvement. Should they be higher? These are pretty old drivers.

Switched back to the first computer (see comment above) and I'm still getting rtapi realtime delay errors (was not getting these errors when it was connected to the mill computer with a base thread of 21000). My max jitter from the latency test was about 25000, but that value did not work, and neither does 35000 where it currently is set at. Maybe I'll run the latency test for an hour or so and see what that gets me.