Unexpected TCP seq number generating connection closing

I have a small farm of web servers running Apache 2.2.15 on CentOS 6, behind a Cisco ACE load balancer, which is behind a Eudaemon FW doing NAT for web clients on Internet and web servers in the farm (and some Cisco Switches between them, of course). Somthing like this:

 __________        ____     ___     ____     ______________
| Internet |------| FW |---|SW |---| LB |---|  Catalyst SW |
|__________|      |____|   |___|   |_ACE|   |______________|
                                                    |
                                                    |
                                             _______________
                                            |Web server farm|
                                            |_______________|

My issue is that I have reports from clients complaining about slow access and/or temporary access errors to Web server (often they have to retry to get access to web pages). When check on server side (the other network elements are outside my administration) I find of TCP resets going out form TCP port 80, and I think this is due to incorrect/unexpected seq number (according to tcpdump and wireshark). Please take a look at following screenshot depicting this problem:

tcpdump capture

Would you please give some ideas about why this might be happening?