MongoDB and Table Storage are not exactly swappable replacements for each other. One is key/value, the other is document. I compared the two in this answer.
There's no getting around the fact that Table Storage is Storage-as-a-Service and you only pay for quantity of data (plus a very small per-transaction cost), whereas to work with MongoDB, you'd either have to host it in your own VMs (which gives you plenty of storage room, but at the expense of VMs) or work with a hoster (such as MongoLab, which offers 500MB for free currently). Regardless, you'd have do do some code changes to work with MongoDB over Table Storage.
I'm not sure if there exists a key/value store equivalent to Table Storage that's locally-installable. No matter what you pick, you'll have modifications on your Azure-side solution if you swap out Table Storage.
Is it possible, for your on-premises solution, to provide a MongoDB backend that stays relatively simple? That is: Stick with a single index to substitute for rowkey, and then store your table entities as documents (avoiding sub-documents)? This would keep your data layout very similar. At that point, you could use things like Aggregation Framework for a bit of data processing, and not damage the overall layout style/schema of your data.
