[jcrawfordor]

Most of the queue would have been with approach, not tower. The aircraft sequenced to land were split across at least two controllers, tower and approach, and possibly more than one approach controller depending on how SFO splits them up. I would expect approach to be able to provide a reasonable estimate but it seems like in this case the estimate was found to be four minutes off, which seems totally within the bounds of a reasonable estimate from an approach controller.

The traffic pattern (US term) isn't really a factor here either way, airliners flying visual still usually use charted routes (like the instrument procedures) or radar vectors rather than the pattern. Remember that the pattern is only about one mile out from the field. By the time airliners are that close they're probably cleared.

[wtallis]

> but it seems like in this case the estimate was found to be four minutes off

No, I don't think you are being at all accurate about characterizing that; maybe you lost count of the delays. After being told to wait ten more minutes, the pilots waited an extra four minutes before pressing for an update, and got told to wait another 10-15 (or f off).