JayBonci is right -- something like mySQL is the best approach. But you say you were asked to "build an output file which combines all the information in all of the input files", so in that context, I think your idea would be hard to improve on, so long as you're careful to use an appropriate delimiter between the fields on each line (i.e. something that doesn't show up as data within any of the fields), and have a sensible way to deal with gaps -- a blank field in a given column for a given row.

Also, if the patterns in different columns tend to be "confusable" (e.g. a search pattern for one type of field happens to match values in another field), you'll need an enhanced "grep" process that can limit the search to a particular field.