I suspect it is not being detected as a failure - but as a planned disconnect. Simply shutting down a VMX will cause a planned disconnect and the client is not likely to failover.
How are you creating the test failover case?
Another option you could consider using is OGS (Optimal Gateway Selection). This causes AnyConnect to use the fastest responding VPN terminator to the user.
This is an option you can configure in the AnyConnect profile. You can also configure a backup server for each of the servers configured as well.
https://community.cisco.com/t5/security-knowledge-base/anyconnect-optimal-gateway-selection-operatio...