A discontinuous fill uses the "Fill replace" command to fill an area.

The Discontinuous fill problem can be illustrated:
Create a 14x14 array of command blocks centered on a chunk at y=0, filled with a scoreboard operation which adds one point to a scoreboard objective. Then create a layer of stone at y=1, then another layer of command blocks with the same commands. Repeat this until you have several layers of "plates" (5-7) Ensure that there is 1 block of air between alternating layers of plates.

...
commandblocks
stone
command blocks
air
command blocks
stone
command blocks
air
...

Your stack of plates should fit within 1 vertical chunk, just to be safe. Now create a discontinuous "fill replace stone redstone_block" across your entire assembly. Count the "correct" number of scoreboard operations. Your score should match this.
The problem is, you'll see a score different from what it should be. gerrybrano, Eease, and myself found out, it seems like only the outermost "skin" of the fill updates, up to 2 blocks deep. This isn't a problem for "planar" layers of command blocks because the fill is only 1 block thick in one dimension, so every command block adjacent to the fill gets updated.
On larger/thicker fills, only those command blocks within that "skin" will have updated and the score will be much lower than the number of operations.