It's great to hear you are looking to build out this scenario. In a topology with multiple shards (hopefully) sharing the same schema, ideally most queries would be contained within each shard. However, in some cases a query spanning shards or relating together data between a shard and a common shared database (containing global data) is necessary. That's somewhat painful today. To the degree realistically possible, not having to write application-tier code to manually handle what most would consider a data tier concern would certainly make the Elastic Scale solution a lot more compelling.
For what it's worth, the other "pain point" of Elastic Scale is schema management. My hope is that existing solutions (SSDT, EF migrations, etc.) can gain "awareness" of Elastic Scale, and enhanced to support cross-shard schema changes. After all, every time "yet another schema management tool" (YASMT) is developed, a higher power sheds a tear!