Memory leakage and degraded performance over time

[rnewman]

@joewalker just reported this. His response time drifted up from ~400msec to ~5000msec over a few days.

Obviously we should point some performance tools at this, but if I had one guess:

Every transact yields a new DB, swap!ping it into the connection. Every tx-report holds on to the previous and subsequent DB. If there's any link between the DB and its last report, then we'll retain every DB instance back to the first. So I'd start looking at the connection and the DB, and work backwards from there, see if anything like that is going on.