one more area we can improve in Mavibot (the profile session I did was
using Mavibot, which is 3 times faster than JDBM).
When we do a search, we will check if the element is present in one
index. As soon as the index grows, we will have to go down the BTree,
fetching many PageIOs. At this point, the root page will be read for
*every single search*. We don't have any PageIO cache in Mavibot atm.
This is most certainly something we need to implement to incrrase
greatly the Mavibot's performances.