A Lucene index is basically an Inverted Flat Index. This means that when Lucene indexes the text for a resource we are interested in, it creates an internal representation where it registers every term found, the number of times it recurs, and in which documents it is found.

So, the real internal structure for an index is somewhat similar to the following diagram:

The structure is what is generally called an inverted index, and explains why Lucene is so fast at giving results in complex full-text searches and at creating and saving indexes, and generally has a limited memory footprint. This structure suggests that once a textual value has been analyzed and its frequencies and positions are saved, we don't necessarily need to save it while updating a field instance in the index. This is why Lucene generally retrieves documents by searches very quickly, but can require a certain amount of time for a full update of very big indexes.