In the event that the number of connection acquisition requests vastly outnumbers the maximum size of a pool, it's possible for the pool to get stuck thrashing between pushing new requests on the queue and popping older ones off when they time out, rather than making any progress towards actually servicing requests. We should consider outright rejecting work when the queue reaches a certain length, possibly as a factor of the maximum size (e.g. 1,000x the pool size).