There are two basic control structures for chemical plants: fixed feeds and fixed products. In the former, liquid levels are controlled by manipulating flows leaving individual units down through a series of reactors, tanks, distillation columns, etc. In the latter, which is called "on-demand" control, the product flow rate leaving the last unit is fixed (by a downstream user), and liquid levels are controlled by manipulating streams fed to each individual, unit sequentially from its upstream unit. Buckley called these structures "material balance in,the direction of flow" and "material balance in the reverse direction of flow". This paper points out that the on-demand structure has several inherent dynamic disadvantages compared to the more conventional approach of setting the feed streams to a process. The most important disadvantage is that level control loops typically have more dynamic lags with this structure, which makes level controller tuning more difficult and degrades dynamic performance. A second problem is that this structure can produce interaction between level and composition loops that is not present in the conventional structure. Another disadvantage is that the on-demand-structure tends to produce larger changes in process time constants as throughput changes, making the plant more nonlinear. In addition, the propagation of disturbances in the on-demand control structure is more complex than that in the conventional control structure because flow disturbances move upstream while composition disturbances move both upstream and downstream. Several of these features are shown mathematically using a simple stripper process. Then they are illustrated on two processes of increasing complexity. The first is a binary system with the reaction A --> B and a plant topology of one reactor, one stripping column, arid one recycle stream. The second is a ternary system with the reaction, A + B --> C and a flowsheet containing one reactor, two distillation columns, and two recycle streams. Dynamic simulations demonstrate that the on-demand structure introduces larger disturbances into the system, which results in more variability in product quality.