Thomas, nobody's suggesting that the caching mechanism should automagically detect that data changed, nor are we suggesting that the cache should be disabled by default.

What Nelson asked for is quite reasonable: allow per-query cache control. When your data comes from several sources where query execution time and data time-to-live vary wildly, there's a good case for allowing cheap, short-lived data to be cached for a short while (or not at all) all the while caching long-lived, expensive datasets for longer. The database table-based cache we're using in this particular case isn't germane to the argument.