We develop an R-tree node restructuring algorithm
that performs post-optimization of existing R-trees
and improves current dynamic insertion algorithms.
On realistic data and relative to state-of-the-art R-trees,
our post optimization technique improves 
point query performance by 36% -- 43% on average (and up to a factor of 2 in
some cases), while only incurring an optimization
cost (measured in disk I/Os) equal to STR loading.
When used to modify existing dynamic insertion algorithms,
our technique results in trees that require
between 25% and 45% fewer disk accesses on average (up to 70%
in some cases),
relative to R*-trees and Hilbert R-trees, respectively,
at only an additional 10% 
insertion cost.