Dimensions Slightly Off

That looks like a CAD\CAM issue.

Can you post the gocde file ?

Also I note that the hal file is not using the "reset" mode
which will half the possible step rate. The hal file is quite old 
(LinuxCNC 2.7 vintage?)

Current stepconf uses reset mode, not sure when this was changed
Note that with you 51000 ns base thread and not using reset mode
the step/step jitter is relatively huge at full speed, that is step to step
at full speed might be ~100 usec one cycle and 150 usec the next

Another possibility is that if the error follows the axis and not the drive,
there is something funny about the X step parallel port pin, or the signal
path is marginal some how.

Gcode files:

Longways y oriented:   

Longways x oriented:   

-same issues happen on 1/4 and 1/8 - I can hear the steps skipping
-not sure what you mean by your second point
-lower acceleration seems to not change anything

machine images attached

A couple other things to note:
-I'm getting the same problem when I use HSM (solidworks cam add in from autodesk, similar to fusion cam) and Cut2D from Vetric
-If I orient the long direction in the X axis or the Y axis, it seems to skip more steps moving in the positive direction (X or Y). The most positive side of the pocket will have that stepping problem between cutting layers as shown before, but the most negative side of the pocket seems to stay lined up 

As a test, can you try setting the maximum velocity to
say 1/2 of the current values to see if that stops the step
loss?
 

I couldn't find much on reset mode in the LinuxCNC documentation - could you explain what you mean? I am using LinuxCNC 2.5 - I didn't install it, but maybe a newer version would work better?

I noted this in another reply, but it follows the longways direction no matter which axis it's oriented towards. Is it having trouble with long travel? I also noted that it seems to have trouble only while moving in the positive direction.

Another thing to mention is that when I home the machine and manually jog it to each limit, it reaches each limit spot-on (no collisions) with seemingly no skipping.

That X axis seems to be a terrible design, from what I can see.

Is there any chance that you can show a picture of each end of the x axis ?

Generally at the driven end of the X axis it's good to see some some of lateral constraint (fixed bearing) and a floating one at the other end.

700 mm/min vs 350 mm/min:
 
Definitely an improvement...does that point to a specific problem? Currently running it in the opposite direction to see if it's better in that orientation too.

You can see where it skips at 700 mm/min, both in the picture above and this picture:  

Like I noted above, it seems to move just fine when manually jogging (at 3000 mm/min) without skipping. It really only skips when running gcode.

X axis attachment images attached.

Looks like maybe a latency problem.

LinuxCNC 2.5 is really old (2012) and may not have had the reset option
If the PC is of similar vintage, it may have issues. Overheating
or hard drive failure can cause latency problems to develop
in a system that worked previously

Since this is such an old LinuxCNC installation, did the machine work ok
and just fail recently?