A polarizing beam splitter doesn't do any irreversible thing (in some language, this would be said as "the polarizing beam splitter doesn't do a measurement"). So if you combine the two beams that come out of a polarizing beam splitter again (actually, using a second polarizing beam splitter used in reverse), then indeed, you restore in principle the exact original beam, with all its entangled properties and everything. Point is, you won't know which path the beam took after the beam splitter (if you do, then that means you've irreversibly altered something, and then you cannot restore the original state).