We need temp-tables to be faster. Apparently the internal behavior of READ-JSON will use transactions (or SUB-TRANS) even if the static data is declared as NO-UNDO.
Since speed is essential when dealing with large datasets, please give us a way to use READ-JSON without the unnecessary overhead. I suspect READ-XML is probably affected as well.
The KB says that this transactional overhead does not affect the IMPORT'ing of data into a temp-table from a data dump file. It doesn't make sense that READ-JSON would have the overhead when IMPORT does not.
More information about the current behavior of READ-JSON can be found here.