Although I'm not able to offer a solution, I can share my theory as to why you are experiencing this seemingly strange removal issue....based purely on my observations and experience with printing with rafts.
Stepping back from the problem, lets think about what we want rafts for. Well, its first job is to provide a level base onto which the printer can start printing an object. Secondly, it has to provide sufficient "grip" to not only keep the expanding object stable whilst its being printed, but also keep the object from curling up as a result of shrinkage forces. And finally, contradicting the second requirement, we want it to easily separate from the printed object once the printing has finished.
The number of layers in the raft is the main factor that allows the first requirement to be fulfilled. If the build pate is off-level or bowed, the first raft layer will vary in thickness to accomodate the varying gap. A second layer is usually required to give a consistent surface for the object to sit.
The second and third reasons, which are contradictory, leave us with having to make a compromise between the first layer of the object adhering to the top layer of the raft, but then not adhering too well so that the raft can be peeled away.
Going back to the reported issue. Based on my own printing, and on the above photograph, I have not seen any direct evidence of variance in the adhesion between the top layer of the raft and the first layer of the object apart from where an occasional stray strand gets included in the raft as its printed. What I have had happen...and what I can observe in the photograph...is differences in the way in which the raft layers peel away from the object when peeled away in differing directions. I'll explain how I see the mechanism of peeling.....
When you peel the raft away "with the grain" of the top layer (ie exactly top to bottom in the photo), you're attempting to roll each strand of the top raft layer away, and at any one time the "width" of the tear is a fraction of the diameter of the strand. Now, when peeling "cross grain", you are attempting to tear a width that is the length of the strand attached to the object. As the accumulated "width" of tear between the top layer and the layer underneath is substatially less , then the tendancy is for the second layer to break away before you have put sufficient pull on the top layer! Peeling away at angles between the two tends to give varying results....
Detail in the bottom photo support my theory. Look at the 5 o'clock position on the ring. There's a hole where the raft has been ripped away at approx 45 degrees in the direction bottom left to top right, where you can see short bits of the top layer have separated from the second layer. With a circular object, the tendency is to lever the raft off from the outside all the way round. Top and bottom, the tearing force is mostly "with the grain" of the top layer....and left and right is "against the grain".
So, if I've identified the problem....how do we solve it? And as far as I can see, I don't think there's a definitive answer.....other than peel carefully "with the grain" of the top layar as much as possible, and finish off with a sharp knife.
Of course there's a possible software answer....to have control over the direction and pattern of the way in which the top layer is laid down. But I daren't even think about how to even define this, never mind the software guys having to program it!