If you are using a drive as an indexing drive for something like a tool turret or spindle orientation then if disabled and moved the drive will keep a not of the encoder and when you command a further index it will move to the correct position but not until a move has been commanded. Or at least that is my experience from my drives.
Hood
Spunk: you are more or less correct in your intepretation of the question. Hood has provided a partial answer for at least one behavior.
Mach is irrelevant, I think, for the very reason correctly pointed out by KTM, i.e. open loop operation. With a true closed loop setup, there is no issue, I would think.
If my understanding is correct, Mach only output's pulses, not positions, so it it up to the drive to determine and maintain the current absolute coordinate.
If you try to move an active servo drive, it sees the encoder moving and applies power to move back to where it thinks it should be . . by reducing the error. In the case of a drive which is 'disabled' (i.e. no motor power) and yet continuing to track the encoder, this error could become great. The concern in implimenting such a feature in an open loop system like Mach is the behavior to expect when the drive powers up and does what it does; which is to reduce the following error to zero, which as Hood pointed out could be a good thing or a bad thing depending on the situation.
So far we have Hood's drive keeping track of the error, but not acting on that error when re-enabled until new pulses are recieved form the CNC. This seems like a sensible solution. Alternately, if the drive does an immediate full power move reduce the error to zero, it could be nasty.
So are there any other behavior that anyone can share? Some drives fault and continue to track the encoder also. So if it knows it has been moved during a fault or while disabled, the reponse to being re-started (but not reset) would be whatever the firmware says to do. I have no experience with drives that have this capability, hence the question.