Transforming ascii file (pseduo database) into proper database

[p.]

I need to take a series of ascii files and transform the data
contained therein so that it can be inserted into an existing
database. The ascii files are just a series of lines, each line
containing fields separated by '|' character. Relations amongst the
data in the various files are denoted through an integer identifier, a
pseudo key if you will. Unfortunately, the relations in the ascii file
do not match up with those in the database in which i need to insert
the data, i.e., I need to transform the data from the files before
inserting into the database. Now, this would all be relatively simple
if not for the following fact: The ascii files are each around 800MB,
so pulling everything into memory and matching up the relations before
inserting the data into the database is impossible.

My questions are:
1. Has anyone done anything like this before, and if so, do you have
any advice?
2. In the abstract, can anyone think of a way of amassing all the
related data for a specific identifier from all the individual files
without pulling all of the files into memory and without having to
repeatedly open, search, and close the files over and over again?