This issue concludes my task for GSoC 2009
The idea is to make all futures from Bigtable paper and other ideas in this processes of to flushing, compacting and splitting.
Some talk in IRC related to this issue creation:
[22:37] <jgray2> bigtable has an optimization which, when flushing the memcache, will actually write out a new storefile that merges the existing storefile with the memcache
[22:37] <jgray2> that way you don't have more than one storefile
[22:37] <jgray2> which is a huge inefficiency