Woot.com and the economics of advertising attention

Lately things that start as an interesting question have turned into something more than that.  A blessing, probably.  Original thought is nice. 

Today, knickknack and 1-day closeout e-tailer woot.com is running a woot-off; instead of the typical woot.com sale, where a single item is available for purchase until it runs out, a seemingly endless stream of woots, with each sold-out item replaced with a new item, of which there is an unknown-or rather, undisclosed- but limited quantity available.

When an individual woot gets close to being sold out, I imagine server load runs to a peak, because users frantically attempt to refresh the page, in the hopes of being one of the first to see the next woot.  Frequently, under these conditions load times for the woot.com homepage  rise, and users are left smacking their F5 keys or hitting Refresh on their browsers, but nothing happens.  Woot's servers clearly cannot easily handle the peak load.

But do they want to?
This is an interesting question.  I bet there is an advertising ROI trade-off between being able to serve up more ads because you can keep up with people hitting reload every .5 seconds, and keeping them queued up so that they stare at ads like "Point Break DVD one day sale" with extra interest and longing.

One thing that makes ad time valuable is if people are really interested in the product/ad vehicle- the ads purportedly have a bigger effect.  I postulate that if you could create some kind of irresistible content like the woots and measure things about them...but it probably depends on lots of factors.  i mean the page itself ius static- once you see the next woot you wiull stop reloading.  I wonder what the situation on the Amazon gold box ads (well, the stuff on the page that wasn't the gold box) were like in terms of clickthrough.

I have seen this problem from a different angle with a video distribution platform
we have modeled- how much server capacity do you need in order to serve your users- which is a factor of each user's acceptable wait time in a queue (and the likelihood that they will abandon the transaction entirely), and the profitability of each transaction.

I think modeling this would be really interesting.