In case you have already done that, please post your load-balancer xml, so that we can help debug your issue.
Health-check is always done for unhealthy instances. This behavior is fixed and not dependent on any configuration. However you can have active health check to figure out whether instances which are marked healthy are actually healthy or not. By this proactive way, you can avoid failures when requests are being sent to instance which is done. However it has some cost associated with it.
Hope this helps,
[Message sent by forum member 'kshitiz_saxena' (kshitiz_saxena)]