Inaccuracy

As I couldn't get the "select edges in empty faces" to work properly but others did, I wondered why is that. So I played around a little and found that this must be due to some inaccuracy of the software.
I noticed it as follows: Create a simple cube. Select a vertex. As the absolute coordinates manually type in those of the position of another vertex so their transforms are identical! Then under 'Selection' in the main menu bar choose "select edges in empty faces" while in edge selection mode. Nothing happens.
The problem also occurs, when I select a vertex and snap it to another while constrained to a dimension (e.g., XY-plane). It doesn't seem to occur, when I move it by hand with no constraints (clicking and dragging the origin of the manipulator). That's also when I can spot a slight change in the shading of the adjoining faces.
So this could cause trouble, as at least one tool already doesn't work reliable, because of this.

I have raised the threshold a bit. See if its better now.

The threshold for the "edges in empty faces" is now way too high. To raise it a little bit seems like a good idea though, just not this much.
I think the problem isn't really with the selection tool. As I described, it seems that sometimes vertices that are supposed to overlap, aren't somehow at the same position. When having a simple cube (1x1x1), look at the coordinates of one of its vertices (e.g., 0.5,0.5,-0.5), then select any vertex next to it, but for this example I'll take the one at -0.5,0.5,-0.5 and in the direct input field now enter 0.5 for the x value. The two vertices should now effectively be at identical positions in space, that is 0.5,0.5,-0.5. For the sake of showing the symptoms now activate vertex snap and click the manipulator at its origin, do not constrain the movement to any dimensions. Snap it to the vertex that originally is at 0.5,0.5,-0.5. You'll notice a change in the shading of the adjoining faces. The values for XYZ however don't change.
This is what made the "edges in empty faces"-tool fail in the first place. Only after snapping to the vertex without constraints solved it.

If you enter the same values for all axes by hand, you'll eventually get the vertices to overlap exactly. The inaccuracy as I call it is also caused, when snapping (vertex/grid) to a point by constraining to 1 or 2 dimensions.

It doesn't have practical relevance at this time, but I'm worried, that it could cause trouble further down the road.

There is a precision issue. The vertex can't get to the exact same position by the two different operations. The normal computation result for the so narrow face is unpredictable. It can be zero. It can also be a vector pointing to any unpredictable directions.

Edit: There are bugs in the new implementation. That's why you see the threshold is too big.

The bugs are fixed.