Excellents tests Tim
Thanks.
They fit pretty well with what you'd expect, and also with my own slightly different tests below. Basically that the environment influences the guide number and that when shooting outside with flash you lose about half a stop.
Anyway, here's what I did - Canon 580EX (identical to the MkII in terms of flash output) and Sekonic flash meter. I tried it outside where there exposure was down on the guide number between 0.3 stops (105mm) and 0.6 stops (17mm with wide panel). In line with Tim's findings. This difference between readings according to the zoom head settings is to be expected as the wider the angle of the flash coverage, the more the environment influences the guide number.
In a 'normal' room, the guide number exposure seemed to be spot on as far as it's possible to measure/estimate, at all zoom head settings.
Moving to a small bright room (bathroom) I was getting over exposure based on the guide number, by 0.3 stops at 105mm and 0.8 stops at 17mm. Here the small room is reflecting loads of light back to boost the exposure, and with the wide panel down at 17mm that makes an even greater difference.
They are in line with what I said above about the environment influencing the guide number, but they are not so much influenced as I suggested. That is, when used outside, about half a stop of the light contributing to the claimed guide number is lost. When used indoors in a 'normal' room, the guide number calculated exposure is spot on, and when used in a small bright room (my bathroom) there's a gain of about half a stop over the guide number.
TBH I think the only significant thing to know is that when shooting outside, you're going to run out of absolute flash power about half a stop earlier than the guide number might lead you to believe. But at all other times of course this doesn't matter because as the auto-TTL takes care of it and adjusts automatically.
General comments, I was impressed with the honesty of the guide number on the Canon 580EX. It's certainly true that not all guns deliver on their claims as accurately, but then again I guess this is a top of the range £300 flash.
Less impressive was the evenness of the flash coverage at wide angle settings. TBH at 24mm it was pretty terrible with a pronounced hot-spot in the middle. I didn't measure it but visually it looked like about one stop down towards the edges and two stops down in the extreme corners. Having said that, at least some of this darkening is down to the lens (24-105L) as all wide angle zooms vignette at short focal lengths, so the error is compounded with flash. Flipping the wide panel down evens this out considerably, but then you lose a lot of power that way.
Things got much better at longer focal lengths and past 35mm it was noticeably better and at 105mm the flash coverage was effectively perfectly even all over the frame.
The lesson there is if even flash coverage is important, take the flash of the auto-zoom mode, which automatically matches the zoom head to the lens focal length, and set the zoom head manually to a wider setting than the focal length of the lens.
) as it really suited the kind of stuff social/party shots those cameras were so good for.
I use mine mostly for testing this kind of stuff! 
