I did not realize that you were using append from. In most of the import routines I have used I used low level file manipulation to scan the file. I have not imported tables of this size this way though. Have you tired it this way?
Low level scanns can be very fast. Combined with the fact that you can take advantage of Insert Into (which is faster than append, not sure about append from) and you won't have to do the global replace statement. Then you can easily implement the n +1 ID scheme.
Just an idea.