I am surprised Veeam blamed this on HA monitoring as this has nothing to do with snapshots, CBT, etc. As suggested in other posts, this is probably due to the VM being very busy from an I/O perspective; thus, it is taking a very long time to remove (commit) the snapshot created by a backup job. VMware is constantly looking for ways to improve VM snapshots. In the meantime, I suggest an agent-based approach for backing up very busy VMs like SQL Server. Many backup vendors tout the fact that they are 100% agent-less. This can be a shortcoming, as we've seen in this thread. Agent-less is fine for the large majority of VMs out there, but certain workloads are best backed up with an agent. Take a look at vSphere Data Protection Advanced, which performs VM backups without the need for an agent, but also includes agents for those exceptional workloads such as SQL Server and Exchange.
vSphere Data Protection Advanced: Backup and Recovery for vSphere environments