Notes:
The most clear advantage is that we achieved our goal of reducing the number of queries on the DB server. And we gained it in a way that can be turned on and off by only changing a single constant! Also, we got there by only updating 17 lines of code and without adding any new hardware.
Because each query was no longer competing for CPU time on a loaded server, the queries could happen faster. But as our data became more complex, we eventually found that we were no longer satisfied with the speed of DBD::AnyData, and had to move back to Postgre (but with replication).
Another drawback of the system is that since we were caching a copy of the data at compile time, any applications that were running when the data got changed would need to be restarted before they saw the change. This was particularly an issue with Apache servers running mod_perl.