This avoid several unnecessary parsings per HTTP request, since in most case
rset's rql is in the rql cache (since it has just been executed), while yapps
based parsing is known to be slow.