Take tiling for instance, if your trying to store a map and are storing a few tiles only across a map, can't you compact that map into a smaller space to be stored for us in the game as a resource by having ints hold multiple peices of information vs multiple ints for the same information? Just curious.
Of course. Like I said, SuperPackME is an excellent example of this. It reduces the number of colors in an image down to the absolute minimum necessary, then stores it in a bit-packed format. That bit-packed format is converted to a hex-string which is inserted into the class data. The resulting class compresses extremely well.
To extract the data, the hex codes are converted back into nibbles. The graphics data is then twiddled out of the nibbles to produce a list of pixels and their colors. The process is described in more detail here
As for someone who stashes levels that way, I think Kevglass's rendition of Tower Toppler might have packed data at the bit level, but don't quote me on that. Many (most?) developers use procedurally generated levels to avoid the overhead of storing level data.
Does that better answer your question?