Poor performance with lots of small queries

[daniels220]

In our application using ReStore, tasks that require faulting in a lot of objects with individual queries (rather than as part of recovering a collection or a satisfying: query) are 2-3x slower in Pharo compared to Dolphin. The profiler says all the time is being spent in Socket>>waitForDataFor:..., i.e. waiting for a response from MySQL—but clearly the problem is not the performance of the queries in MySQL itself, since Dolphin is using the exact same queries, just issued through its ODBC connection. Unfortunately I don't know enough about the internals of either ODBC or the MySQL socket protocol, nor the Pharo sockets implementation, to have any idea where to start finding the actual cause of the problem.