Strange Problem with 10Gb Ethernet Blade Switch for c-Class BladeSystem

[ColTech]

We have had a problem for a long time now that can be summarised as this:

If we route traffic through our 10GB blade switches (either one or both load balanced) we experience a high latency on database based applications e.g. sharepoint, exchange 2010 etc..
Routing the same traffic on the same VLAN through the 1:10GB switch massivley changes response times e.g. from ~100 ms RPC times in outlook to ~5 ms RPC times.

This effect can be observed in VMWare (ESX 4.1) and blades with server 2003 64bit OS loaded on them. Bandwidth does not seem to be a problem more latency. Ping times are normal and no packet loss.
Wireshark does seem to show lots of TCP DUP ACKs when running through the 10GB switches.

We have changed uplink transceivers and uplink fibres. We are uplinking both the 1:10 and 10GB switches to an 8200 series procurve switch with OM3 cables.

I am now at a loss , can anybody help?

Many Thanks in advance