This is caused because you are most likely redirecting the headset from the client to the VM. This results in the following problems:
1) media hairpinning
2) high bandwidth upstream from client to remote desktop (the audio is sent as raw usb data instead of encoded/compressed audio)
3) higher load on VM
It is not recommended to use a VoIP application in this manner - especially in a call center environment. In a call center, VoIP quality and SLAs are extremely important.
We have the SE SDK generally available. We have UC partners that have used this SDK to re-architect their VoIP application to work optimally in a VDI environment. More info here.
VMWare Interactive Media Services/Session Enhancement SDK Access - Partner Communication
Public SDK
The VMware Session Enhancement SDK is now a publicly available SDK on the VMware communities website - http://communities.vmware.com/community/vmtn/developer/forums/horizon_vse_sdk