Hm, I get this working with postgresql by hacking indexer_rdbms.py so
that it first finds issues containing each of the words, then intersects
that.

Here's some example generated SQL for AND'ing at issue level, if it's
too stupid please enlighten me :) :

select nodeid from issue_messages where linkid in (
select CAST(_itemid as Integer) from __textids where _textid in (
select distinct(_textid) from __words where _word=E'NICE')) INTERSECT
 select nodeid from issue_messages where 
linkid in (
select CAST(_itemid as Integer) from __textids where _textid in (
select distinct(_textid) from __words where _word=E'TWO'));