We show that simple kinematic arguments can give limits on variabilities in gamma-ray burst (GRB) afterglows. These limits are violated by X-ray flares in the early afterglows recently identified by the Swift satellite. We discuss that a probable solution is that the central engine continues to eject an intermittent outflow for a very long timescale up to 1 day. This long-lived engine model may also explain the flat decay of early X-ray afterglows, while the gamma-ray efficiency of GRBs should be incredibly high (>75-90%) in this model. We suggest new possible models to evade this efficiency crisis and discuss implications for future observations.