Closing the circle:
RethinkDB gets around these issues in the following way. We identified over a dozen parameters that affect the performance of any given drive (for example, block size, stride, timing, etc.) We have a benchmarking engine that treats the underlying storage system as a black box and brute forces through many hundreds of permutations of these parameters to find an ideal workload for the underlying drive.
Original title and link: RethinkDB: On TRIM, NCQ, and Write Amplification (NoSQL databases © myNoSQL)