I don't know NAMD very well at all, but I assume you're using periodic boundary conditions, in which case this isn't a problem at all. In periodic boundary conditions, each molecule you see represents an infinite number of molecules, one in each of the infinite unit cells. Let's look at this in 1D, since it's easier to draw. You see this:

| X |

where X is a molecule, and | is a boundary of your unit cell. The actual system is the infinitely periodic 1D system, i.e.:

... | X | X | X | X | X | X | X | X | X | ...

Now, let's imagine that the molecule "leaves" the unit cell by crossing the left-hand boundary. You will see a series of frames looking something like this:

| X |

| X |

|X |

X |

X| |

X | |

It looks like the molecule has left the unit cell, which shouldn't be possible! However, this is because we're only displaying one particular choice of the (infinitely many) molecules. If we look at the same trajectory, but putting all the unit cells and molecules in, we see:

| X | X | X | X | X | X | X | X | X ...

| X | X | X | X | X | X | X | X | X ...

|X |X |X |X |X |X |X |X |X ...

X X X X X X X X X ...

X| X| X| X| X| X| X| X| X| ...

X | X | X | X | X | X | X | X | X | ...

so as the particular molecule you were looking at leaves the left edge, its periodic image enters on the right edge, it's just that your display software is choosing to show the original which has left the cell, and not the periodic replica which has just entered. Some software will make the opposite choice and any molecule leaving an edge will disappear and be replaced by one entering at the opposite edge -- however, this discontinuous motion looks very odd, and can make it very hard to track things like diffusion where you might want to know how far a particular molecule moved.

Hope that helps!

Phil Hasnip