Unfortunately that's a bug and you should ask your sales rep to get you a WAN eval license for 7.0.
Something else you may also consider is a slightly different approach with 6.6
Essentially the setup looks like this for each server:
Create a local region with a write-behind gateway associated with it. This implies that you would only ever have a primary gateway and no secondary as the region is local.
Create your normal replicated region and attach a CacheWriter which updates the local region (containing the gateway).
This would result in every server having it's own local region (not visible to anyone else) and a unique primary gateway. With this setup, events would be able to be processed in parallel. This setup does have a couple of caveats, however:
If a server goes down and never comes back up, you lose the events in flight as there is no redundancy for the queue in this setup. If the server comes back up, you'd be OK as the gateway queue is persisted to disk and processing would be able to continue.
The second caveat is that events would be processed asynchronously thus there is the potential of processing events out of order. In your existing setup, two updates for the same key are guaranteed to be processed, by the gateway, in the correct order. With multiple gateways there's no guarantee that the same gateway would process the same key. To help with that, you'd definitely want to use a partitioned region so that the same key is always processed on the same system. This still doesn't help when you're updating different keys and require update ordering, but it may be enough depending on what you're doing.
--Jens