tedster - 2:42 pm on Aug 16, 2012 (gmt 0)
Trying to mind read Google here (not always the wisest thing) I do think that handling user's query ambiguity may be part of the picture. After all, the "ten results per page" layout is just a long standing convention, grounded mostly in the fact that our western number system is base ten. It's just fingers and toes, not solid testing.
For generating a main menu, I've tested the number of choices I throw at visitors pretty extensively - and six to seven choices seems to be the sweet-spot and also a break point. The minute the menu hits eight choices at the same "level" of logic, the number of bounces usually goes way up. I always say "too many choices might mean no choice at all."
When the US migrated to longer phone numbers, Bell Labs did extensive testing on how long the new numbers should be for the sake of user memory - and seven was their result. A menu or a SERP has a similar challenge. To make a choice, people need to remember and quickly compare what's being offered - so I could see seven being a sweet spot, especially for the challenged search user who makes an ambiguous search.
A few months back people were spotting a lot of experiments with fewer than ten results on Page 1 - as low as four. However, this looks like it's not a test. As you said it may be a bug, but it seems to be too wide-scale to be a test at the pure research level.
Even when the query is branded, Google may see a better total click through with the Sitemap-plus-six configuration. We should remember that their primary purpose is not to give as much exposure as possible to our sites, but rather as much satisfaction as possible to their user - as if anyone could forget that in the midst of this past year's changes.