The reason uncached reads are slower in Cassandra is not because the sstable is inherently io-intensive (it's actually better than b-tree based storage on a 1:1

The reason uncached reads are slower in Cassandra is not because the sstable is inherently io-intensive (it's actually better than b-tree based storage on a 1:1 basis) but because in the average case you'll have to merge row fragments from 2-4 sstables to complete the request, since sstables are not update-in-place.