Yes, they should be the same. But different values should not result in missed steps, only in uneven movement over different cartesian X/Y positions. Lover values are safer but even values of about 10000 should be usable with a delta printer. So you already have it very conservative. I would be surprised if fixing acceleration helps you with the sudden layer shift.
Edit: Stepper driver overheating or a pulley slipping or maybe even too low current on one stepper looks more probable.
Strange that the error is very nice (about the same over the whole top area of the part). Stepper skips (or pulley slides) lead also to Z-height error and not only X/Y error. All the X/Y/Z errors because of that are about the same magnitude. It does not look like your part has any Z-error there.
Edit: Stepper driver overheating or a pulley slipping or maybe even too low current on one stepper looks more probable.
Strange that the error is very nice (about the same over the whole top area of the part). Stepper skips (or pulley slides) lead also to Z-height error and not only X/Y error. All the X/Y/Z errors because of that are about the same magnitude. It does not look like your part has any Z-error there.