This is a classical skip mode operation where the feedback voltage (the voltage at the comp pin) is internally monitored by an extra comparator featuring hysteresis. When the feedback is above a certain level (I believe they say 500 mV in the data-sheet), the circuit switches normally with a full-length switching pattern. When the output current reduces, the feedback voltage also does and the duty ratio goes down bringing the peak current down. At some point, the load is such that the feedback voltage passes below the 500-mV threshold. Because of the extra comparator, all switching cycles are interrupted and the power switch is turned off. See the below simplified sketch for illustration:
When the switch turns off, the output is left with the capacitor being charged at your regulation voltage (3.3 V it seems) and the current absorbed by the load and the divider network. As such, the rate at which \$V_{out}\$ falls depends on the time constant \$C_{out}R_{load}\$: if the output current is truly zero amp (no-load) and if the resistive divider is of large ohmic value, then it can take a large amount of time to reach the hysteresis band and reactivate the switching cycles again (as \$V_{out}\$ falls too far from the 3.3-V target, \$v_{FB}(t)\$ rises up again). With your scope, sync in normal mode trigger, not auto, so that you can see the bunches at a low repeating rate. Another option is to slowly reduce the current from full load until it starts skipping pulses. As the current goes lower and lower, the distance between the bunches expands: the converter is now operating in a hysteretic way. If you want to further dig the world of switching converters, you can check this book.
