Currently, when the table cardinality (number of records) gets estimated, all records are considered to be compressed twice in size (hardcoded compression ratio of 0.5). It would be better to have some estimation (even rough) of the actual compression ratio, because in practice it can differ much from the expected value. This becomes more important starting with versions 2.1.4 and 2.5.0 that have the join cost more depending on the proper cardinality estimations than in prior versions.