Analysis of incremental slip rates from the four major strike-slip faults of the Marlborough fault system (MFS) of northern South Island, New Zealand, provides a first-ever record at the scale of an entire plate-boundary fault system of how relative plate motions are accommodated in time and space. This record, which spans the past 350–450 m of relative plate motion and ca. 12–14 ky, demonstrates that the fault system as a whole accommodates a steady plate-boundary slip rate, with the MFS faults “keeping up” with the overall rate of relative Pacific-Australia plate motion at relatively short displacement (10 s of meters) and time (102–103 yr) scales. These results affirm the often-assumed but until now unproven assumption that the relative plate-motion rate provides a robust basic constraint on both geodynamical models and analyses of system-level seismic hazard at these scales. In marked contrast, the incremental slip rates of each of the four main Marlborough faults are highly variable through time, marked by coordinated accelerations and decelerations spanning 4–6 earthquakes and several millennia as the faults trade off slip to accommodate a steady relative plate motion rate. These results suggest that (a) the weakest fault in the system will slip faster than average while adjacent mechanically complementary faults slip more slowly, and (b) that these patterns switch back and forth through time, likely reflecting reversible changes in the strength (i.e., resistance to shear) of the individual faults as they collectively accommodate relative plate motion. Interestingly, the periods of fast slip on the MFS faults exhibit ∼20–25 m of displacement, suggesting that these may record periods of fast slip on a weakened fault/ductile shear zone that continues until it uses up all locally stored elastic strain energy, thus potentially approaching local complete stress drop, albeit during a few tens of meters of rapid fault slip during multiple earthquakes, rather than during a single event. This hypothesis is consistent with typical earthquake stress drops of ∼1–10 MPa and estimates of depth-averaged crustal shear stress of a few 10 s of MPa, such as might be released in clusters of 4–6 earthquakes. These results emphasize the need to analyze the collective behavior of the entire fault systems, rather than just individual faults, to understand the mechanics of the system. Moreover, these patterns suggest a potential path forward for more accurate estimation of time-dependent seismic hazard, with the possible incorporation of current position of a fault within a fast- or slow/no-slip period into the probability analysis, as well as a means of potentially estimating crustal shear stress.