Difference in distance measurements between encoding directions

[Xen]

We have scanned a cuboid phantom several times using spin echo and fast spin echo sequences in different planes and with flipped encoding directions, and we always find that the dimension of the phantom across frequency encoding/readout is larger than phase encoding (1-2%). When we scanned it using another scanner (different vendor) and the same parameters we didn't see this difference. I know it's an issue with our sequences, but what exactly?