We have found that the Virtual Table interface is sufficient to implement any type of backing store without changes to anything internal to SQLite. Specifically we have operational virtual table implementations for - Faircom CTree files - Our own proprietary Data Dictionary (residing in shared memory sections) - diverse in-memory linked lists in C - FastBit compressed bitmap indexes - structured TLV decomposition What, specifically, would the unique selling point of your proposal be that cannot be achieved using a virtual table?