If you read Microsoft’s documentation and guidance regarding DAG design, you will note guidance that the greater number of DAG members that you have, the better. This is because, with more DAG members, you can increase your resiliency in the event of member failures.
For example, with two, three-node DAGs, each DAG can sustain only a single member failure and still remain online. Since a three-node DAG requires two voting members for quorum, two simultaneous voting member failures would result in a loss of quorum and the DAG going offline. However, a single, six-node DAG (stretched across sites if desired) can sustain three simultaneous voting member failures before going offline. Since a six-node DAG would require four online voting members, you could have three Mailbox server members online plus your Witness Server and sustain simultaneous failures in the three other Mailbox server members without losing quorum.
The caveat to this guidance occurs if you need an active/active data center model. Consider, for example, that you had two data centers (West and East) and wanted half of your users to be primarily hosted in one data center and the other half to be primarily hosted in the other. User allocation could be based on the quality of WAN link to the data center, the assignment of users to other Exchange integrated applications in those data centers (BlackBerry, document management, etc.), etc.
Under normal operations, if you placed three DAG members in one data center and three DAG members in the other, it would work fine and you would have the added resiliency benefits of a larger DAG. However, if the WAN between your two data centers were to fail, all of your mailbox databases would failover to whichever data center contained your configured File Share Witness (if a valid mailbox database copy existed in that data center). This is because, in the data center without connectivity to other DAG members or the FSW, you would no longer have enough voting members for quorum and the DAG would take itself offline.
While having Exchange services failover to one data center or another in the event of a WAN failure may not be an issue, it is technically a site resiliency operation that may not be desirable for end user access, integrated applications, or administrator management. For these reasons, if you wanted each data center to remain active in the event of a WAN failure, you would need to deploy two DAGs stretched across both data centers. One DAG would have its FSW hosted in one data center and the other would have its FSW hosted in the other data center to ensure that Exchange services remain hosted in their designated data centers, even during a WAN failure.
For more in my series on Exchange 2010 Notes from the Field, please click here.