It sounds like the EX90s are not negitiating a content stream.
A few things:
You mention that the Neighbour zone is H323. Are the calls themselves SIP or H323? Have you tried both? Interworking between SIP and H323, or eliminating interworking, can often affect things like content. You have multiple call legs, so it could be SIP - H323 - SIP, or H323 - H323 - SIP, or H323 the whole way for example. Try different combinations.
I'd also suggest trying to allow SIP on the neighbor zone - everything is moving to SIP these days, H323 is considered legacy from Cisco's perspective.
What are the software versions of VCS and endpoints?