This issue I’ve seen with larger SOBRs having many ReFS extents. Having a backup-copy job (pruning) with GFS there were sudden unintended switches of backup chains to other extents. Thus loosing all space efficiency for the GFS points. We used scripts here to detect chains spanning more than a single extent. It turned out to be a bug solved in V10 paired with a very conservative process in VBR to detect extents near space exhaustion. There is some estimation here due not being able to estimate the space savings due to block recloning.
This could be tweaked with two regkeys:
Pfad: HKLM\Software\Veeam\Veeam Backup and Replication
Name: SOBRSyntheticFullCompressRate
Value: 35, DWORD
Description: override the estimated space VM would take on SOBR in a full backup. % of previous full backup size.
Name: SobrForceExtentSpaceUpdate
Value: 1, DWORD
Description: enables advanced SOBR extents free space update logic. With this set to 1 (enabled) service updates cached extent free space every time task is assigned.
Maybe it helps.
Regards,
Mike