F5 LTM sending reset packets to Pool member

I believe you will see this behavior when you have an HTTP profile on the virtual and a problem occurs on the serverside or with the internals (modules/plugins) of the BIGIP system itself. I could see this happening if no SNAT exists, there is a conflict with a module/plugin or the address used for SNAT or the server's SYN/ACK is never returned (whatever the reason).

A standard type virtual server maintains connection state independently. If you have an HTTP profile then BIGIP will not attempt a serverside LB pick until the first HTTP command is seen (typically a GET). This "delayed binding" behavior can be confusing since it looks like the VS is resetting the client but really it MUST do so since it is unable to communicate with pool resources (despite forcing it to be marked up via ICMP).