Normalization Question Regarding Column Combinations

We need to store land title information about properties in various
Australian states, but each state maintains it's own land title
registry and use different columns (well actually different
combinations of the same columns). For example:

Victoria store:

TorrensUnit
TorrensVolume
TorrensFolio

Queensland store:

TorrensCounty
TorrensLot
TorrensPlan
TorrensParish
TorrensUnit
TorrensVolume
TorrensTitleRef


There are 11 different columns and they are used in 8 different
combinations depending on the state.

Since we need to store information about land in different states I see
two possible solutions:

1. A sparse table containing the 11 columns with a CHECK constraint to
enforce the valid combinations.

2. A table for each state containing only the columns relevant to the
state with a foreign key relationship to the table containing the
common columns.


I'm not sure if the data type and length is consistent between states
yet (waiting to find this out) but assuming that it is which of these
approaches is going to be the most rigorous? I'm leaning towards (2)
but I don't like the feel of a table per state.