"I split my data to their own table by list_id field" Does this mean that there are now a whole bunch of tables with identical structure and with the previous list_id as table names? In that case, you would be searching SQLite's list of tables while preparing the statement, thus hiding much of the cost. Table names are NOT data. FWIW keeping the UUID in 16 byte blobs instead of the 36 byte canonical representation would cut the time required to compare them in half, with input and output handled by user defined functions