Shell Programming and Scripting

Are you replacing the tags or the data?
If you are changing the tags, it might be simpler to parse the xml file, create a database, then re-create the xml file using the replacement tags.
Replacing the data is somewhat trickier if the replacement word is a portion of the data field.

I'm replacing data.. :-( If I was replacing tags, I would have loaded it, changed the tags and exported it like you said.

Some assumptions:
If the 100mb file contains 1 to 2 million records, and the control file is 10000 lines at 13 characters per line, using a simple sed routine will process about 130gb of data. Whether this is all disk i/o or memory will depend upon how well the shell uses memory.
If there is only one data field per line, create two temporary data files, one containing the tags, and the other the data. Add line numbers to the files.
Sort the data portion into data portion sequence. Sort the control file into original field sequence, write a merge program to read the sorted data only file, replace the field, and write a new temporary file with the new data (including the original line number).
Sort the new temporary data file back to line number sequence, and merge with the temporary tag file to produce a new xml file.
The total data processed this way should be less than 1gb.