Hi Alexandre,
I made some searches on this same topic some time ago and came to the same conclusion as yours regarding FP indexes. FP indexes are not the same kind of indexes we are used to work with (B-trees).They are not optimized for selecting a subset of rows out of a larger set of rows . However, FP indexes are really fast when rows need to be projected/materialized (join the RowID from several columns to construct a row). As the number of values into a page is constant, when you have the RowID (page#+row#) from one column of a table, it is very easy/fast to find the corresponding RowID into another column of the same table using the FP index. It is just a matter of arithmetic.
Hope this helps.
Marco
PS: These are my personal findings on this topic, and, may be proven to be wrong.