*snip*
> try to create some metrics and base your method on those metrics. mysql
> can and does a good job of either of these methods.

mysql might, but the other layers might not. I believe JDBC has
substantial overhead obtaining a connection when not using connection
pooling, but I may be mistaken.