To add on to Beth's idea, you'll need to open the 4000 files at some point, it may be easiet to read in the 100-300 strings, cache them in a structure (using less memory) and then iterate over the 4000 files - 1 time each - looking for all ~300 strings?