Look into: Indirect indexing of tx data by a hash column

[i-norden]

Tx data is too large to index directly by btree

One potential option is to index it by storing the hash of the data in a separate column and indexing this column, but additional research would need to be done to determine if this is advantageous enough