Splay trees allow for more efficient searching of data sets where particular items are frequently accessed. I think this could help with searching data sets where there is a high degree of autocorrelation. For a fairly good description of the idea refer to the link below (no, I'm not one of the authors, but a CS colleague recently introduced me to the value of splay trees). I know this is probably a case of 'easier said than done' but thought I'd throw it out there.
https://www.google.com/accounts ( https://www.google.com/accounts ) - 2012-06-12 18:48:50 PDT