High Fail rate with MOD_JK and JDG remote-cache-container

Scenario: 2 nodes WildFly cluster where session data is offloaded to a remote 2 nodes JDG cluster (jboss-datagrid-7.3.0);
When using MOD_JK as load balancer we observe a higher fail rate (~9%) than when using MOD_CLUSTER or HAPROXY (~0.9%); fail rate is intended as inconsistencies in the data stored in cache and subsequently retrieved on another node of the cluster:

• MOD_JK
• MOD_CLUSTER
• HAPROXY