The manual and the simulation values show the laser tops out at 133 mm/s on both X and Y. You can set any layer speed you want, but the controller will not run the steppers at more than the maximum speed set in the Machine Settings.
IIRC, GRBL uses mm/min for the speeds shown in the Machine Settings, so you must convert from mm/s when comparing the numeric values.
The 500 mm/s² acceleration means it requires about 18 mm to reach 133 mm/s, which is likely much longer than 1% of the speed will produce, particularly for one-word-at-a-time scanning.
The result is it’s proceeding at a much slower pace than you think. That makes diagnosing the problems much harder, because all our assumptions are wrong.
Those are likely due to USB communication errors dropping out chunks of the commands. Using a better cable, typically one with molded-in ferrite cores, seems to improve the results. The Troubleshooting page has some suggestions:
An earlier discussion suggests how long it takes to recover from various other non-solutions: