What I've seen is that most audits are incomplete, because the audit is time-boxed. The auditors only have a few weeks to examine the books, either because that's all the budget by the client will allow, or because the audit firm hits the profitable/not-profitable limit around the extra services they sold to the firm.
So the firm has an incentive to draw out the process and gamble that the auditors won't find anything actionable before the clock runs out.
This is a very common misconception. Whilst auditors should always report fraud if they find it, an audit is not there to detect fraud. They call this the "expectations gap".
An audit is trying to demonstrate that the accounts of a company show a 'true and fair view' of the companies performance. Things that interest auditors are things like, is the stock all there, will the company be able to sell the stock for what it bought it for, does the income booked to these accounts properly belong in this time-frame, are these assets still worth their book value.
Obviously this turns up various kind of fraud, but wouldn't for instance detect a fraud like Libor.
edit:
> So the firm has an incentive to draw out the process and gamble that the auditors won't find anything actionable before the clock runs out.
Auditors are trained to look out for this, but it is certainly possible to some degree. I would say it is getting harder to do, because much more of the audit methodology is about 'analytical review' (do the results seems possible) rather than 'substantive testing' (checking documents). If the audit partner is not happy with the access he has been given he can always 'qualify' the accounts in his statement. This is then public information, and a qualified audit would raise a lot of suspicion with counter-parties
So the firm has an incentive to draw out the process and gamble that the auditors won't find anything actionable before the clock runs out.