Judging by the error message ("Device or resource busy:"), it seems likely to me that NFS was the culprit here.  The fact that the tests passed when run alone may point to an issue with the way tests were run as a group, but this has changed significantly since 2016.

If this can still be reproduced with modern versions of everything involved, it may be worth opening a new issue.  In the meantime, I'm closing this as out of date.