If there are (or will be) investors willing to invest at the initial IPO price, how can you say that price is going to be 'artificially high'?

The IPO investors, themselves will be buying in the hope of making a gain, if their analysis/forecasts/judgement is wrong - it won't be Goldman that screwed them, it will be failure of their own judgement. And also their own willingness to join to 'bubble' to make a gain.

Also, if Goldman are the underwriters of the IPO, and they fail to sell the shares at the 'artificially high' price, they will have to take on the shares themselves. If such a high profile IPO is not fully subscribed, it won't reflect well on Goldman, so setting an artificially high price, is not in their interest, but setting a marketable price is.

Also empirical evidence suggests IPO's tend to be underpriced as opposed to over priced, to allow for gains on day one. (see http://en.wikipedia.org/wiki/Initial_public_offering#pricing)

With Goldman themselves having $450m in facebook, this also adds some credibility come IPO time, as GS themselves will have 'skin in the game'.

Shouldn't the bank that does the IPO try to set the price as high as will sell anyway?

Not at all- it's similar to the Real Estate Agent problem that is described in detail in Freakonomics. The bank has some slightly higher interest in raising the price and getting more money- but a much higher interest in offloading as much security as possible. This has led to problems like banks selling to specific IPO turnaround firms that can grab huge share volume early on, sending the IPO up in value before selling out right near close of trading, sometimes trashing hot IPOs on day 1.

You can read about this: http://antisocialmedia.net/ipo-probe-of-wall-street-ties-ton...