// more slowly than the other queues (which are using more seats than they
// are allocated).  The implementation has a hack that addresses part of
// this imbalance but not all of it.  In this test, flow 1 can not use all
// of its allocation during the first half, and *can* (and does) use all of
// its allocation and more during the second half.
// Thus we expect the fair (not equal) result