Self-IP from all F5 guests are sending GET requests to the web servers.

500 per 10 seconds is not realistic. That would be 0.02 second interval in between the HTTP health check requests to a single pool member. Interval this short would cause each pool member to flip available/unavailable even in ideal LAN conditions.

What you're seeing is not traffic that originates from F5 BigIP. It's traffic that F5 sends to pool members after normal balancing/routing procedure. That traffic originates from elsewhere. Is one or more of your Virtual Servers configured to use SNAT automap or SNAT Pool feature? That would explain a packet capture and/or server logs where BigIP's SelfIP is seen as the client.