@BartArondson here are two: sandbox requirements can affect apps that read / write and are not updated. Spotlight data detectors process all writes to disk, so if an app writes data that are rush in metadata or call a slow processor, then the overhead can be benchmarked, noticed.
Here's another. System logging, package receipts are all using sqlite database files for much of the writing. If you never log a single message, then that overhead is wasted.
However, if you have thousands of log messages a second, the efficiencies of a database could make the processing of messages far more efficient than the old syslogd manner.
All sorts of system level abstractions have been introduced steadily - so based on where someone's bottleneck lies - the changes could be better, worse or a mix.