To put it in a simplistic fashion:

When the index is badly fragmented, and is analyzed (ie., stats are available), the cost based optimizer will find it too expensive to use the index and go for a full table scan instead.

However, when the Rule based optimizer is used, it might be 'hard-coded' to use the index and therefore index would be used, no matter what. If there are large number of deletes/ updates to the indexed colun, might lead to poor performance?

However, things are not that simple as they seem (or should be). There are a large number of variables interplaying.
Experts feel that there is no performance gain obtained by index rebuilds........