Hey,
I am working on exactly what you are looking for (i am planning to eventually make it open source).

Note that my approach is probebly overkill in your situation, it is meant for large data sets (tested on 250MB of text & works surprisingly well).
I found that the best way is to set up an inverted-index of all the terms as well as an index which shows the position of each word within each document.
I then use an algorithm which gives a bonus if the words that are being searched appear close to each other in a ducument -- this proximity-search algorithm is described at http://citeseer.nj.nec.com/cachedpage/550719/1 .
Also to improve the inverted-index words are indexed by their stem (a stemming algorithm can be found here http://www.ldc.usb.ve/~vdaniel/porter.pm ).
Aswell I have implemented an algorithm similar to google's pagerank (a good description of it is at http://citeseer.nj.nec.com/cachedpage/368196/1 ), the popularity of a page is taken into account when returning results.

I use MySQL for all the storage / indexes.