A falling edge ( off ) signal does not travel through the last repeater in a chain if the last repeater delay value is equal to the pulse length of the falling-edge signal.

Take the example shown in attachment 'Rep-1'. A three tick pulse is sent down two repeater chains (red - falling edge, purple - rising edge). All of the repeaters are set to three tick delay (same as incoming pulse length). The rising edge signal travels through all of the repeaters no problem, the falling edge signal stops at (doesn't even activate) the last repeater.

This is the case with a chain of any length, from two repeaters upwards. It is also the case for any matching pulse-width/repeater-delay pair - two, three and four tick are all affected (I was unable to create a 1 tick falling edge pulse so could not test this case).

Attachment 'Rep-2' illustrates two alternative falling-edge chains that do allow the signal to pass through. Red is the same defective chain as shown above. Green demonstrates that by reducing the last repeater to one (or more) ticks less than the pulse length allows the signal to pass through. Yellow shows that a one block gap of redstone between each repeater lets the signal through.

I consider this to be unintended behaviour, as it is only the last repeater in a chain that is affected. A three tick falling edge pulse can travel through any number of three tick repeaters without being blocked (except by the last).