This is a rock and hard-place for nearly anyone in this position, as I've been there as well. Most recently, I've seen colleagues have to make choices around NoSql key-value stores (Mongo, Redis, Cassandra, Riak, etc.)

I agree with AndrewDucker to a certain extent: structuring your application architecture to allow one to move from one sub-system to another is ideal. While appropriate in generalities, it's the specifics that matter. Most substantive environments don't interact with these systems through a simple connection that can be hot-swapped. There are dependent processes that are often designed around the fundamentals, quirks and intricacies of a given technology.

The scope of one's implementation definitely affects how much you can hedge your bets if you need to shift to another technology stack.