At present, the library's oid manipulation functions are slower when built in SHA256 mode than when not. Add some microbenchmarks around the oid compare and copy functions to understand this better.
