For LEDs which are relatively high current loads switched at logic speeds (5ns perhaps) I'd use 0.1uF + 10uF per chip,
and 100uF+ for the whole board.
Switching many LEDs simultaneously could be causing a rate-of-change of supply current of the order of 10^8 amps/second,
way beyond what a voltage regulator on a remote board can handle, so in a microsecond or so those 0.1uF capacitors are
drained and the circuit malfunctions - before the voltage regulator has time to adjust to the new load (which might take many
us).
The extra decoupling should hold the supply voltage up for long enough.