Apparently, most editors of sci.  journals publicly agree that to some extent they are responsible for publishing fraudulent papers (as well as erroneous, plagiarized etc.). So did Nature's editor P.Campbell, declaring in a recent big "briefing on science and fraud" that:
"Nature's policy is to publish whatever information it can about published papers that have proved suspect or false. Although after due consultations" (Nature, 398:15, 1999).

Yet, of course, officials may lie (I have informed Mr. Campbell about this suggestion) and actual behaviour of editorial boards may significantly differ from announced principles. Here are the actual instructions I received from Nature editorial office in response to my finding proving that some paper published in this journal is at least "suspect"
I suggest that you write to the authors themselves and to their institution and/or funding source (as detailed in their paper) so that the matter can be looked into. If it does emerge after investigation by the funding body or institution that the authors published fraudulent data, the authors will be asked to withdraw the relevant papers. In this event they will inform us and we will print a correction.
That is, I had to bring from the authors of fraudulent publication their permission to publish my findings!!!???

Surely, my case is not unique; take look, for example, at relevant chapter of  a book "A Habit of Lie" by John Hewitt discussing real world editorial practices from more philosophic point of view.   Take look also at Editorial in GUT v.48, March 2001, pp. 285-286, ("Retractions in Gut 10 Years After Publication,") for the story of a whistleblower who successfully went through all procedures prescribed above by Nature editors. I thought it is impossible, but I was wrong!

So, in more details, in 1996 I have published in Internet an article proposing (among other things) a funny "indicator" of sci. fraud - physically impossible appearance of certain type of graphical data presentation which seems impossible to explain otherwise than by "absent-minded" data fabrication. (That my finding is shortly explained at the picture below.)

In 1997 I have found an instance of such graph in Nature and sent a "whistle blow" (appended below) to its editorial office. At first, there was no response at all, even after an additional query. Only after I have sent appropriate information to the sci.fraud maillist, editors vouchsafed me fruitful advice to send my information to anybody else. (Perhaps, it was meant by Mr. Campbell statement that performing "due consultations" is exclusively whistle blowers' burden?) Then it was said that my accusations are "too serious" whereas I have presented no "compelling evidences". And finally Nature published an elegant confession by the authors that something "was misdrawn" in their paper without mentioning my submission at all. (see further correspondence).

I think, it is a quite appropriate behaviour for editorial board of scientific tabloid. Russian proverb ruling it is that "the crow would not peck at the other crow's eye" (I will look for english equivalent, help would be greatly appreciated). Actually, it is also difficult to blame Mr. Campbell for his lie and to pretend that he should call a spade a spade and openly admit this policy.

Perhaps, what I should do in these circumstances is to sue Nature for copyright issues. Indeed, I have found an error in Nature article and published my finding in the Web. So, apparently, I have a copyright on this finding. Nature editors (obviously intentionally) published several months later the same finding signed by different author. I guess, this is piracy.
I do like this very extravagant idea (many thanks to John Hewitt for its original version). It is like suing Holy Inquisition  for breaching copyright if we can not prosecute them just for burning heretics. Very interesting... Yet, of course, I have no money to pay British lawyers.

Moreover, none of Scatchard plots on Figs.2,3 can not be based on any real data. It is a peculiar property of Scatchard plot that both X- and Y- coordinates on this plot are linear functions of measured signal (Bound); so pairs of datapoints obtained at the same concentration of added ligand in two experiments should lie approximately on the line drawn through the start of coordinates. Obviously, these pairs in Fig 2 and 3 lie precisely one under another - it is a rather typical error in fabricating Scatchard plots.

Perhaps, my paper 'Absurd Trivial Errors in Scatchard Plot Analysis' (http://www.orc.ru/~yur77/absurd.htm) may be of use for your 'peer reviewers'. I think it is shameful for such pretentious journal like yours to publish works committing any of those errors. 

Additional graphical illustration (this is NOT the original figure discussed in my letter to Nature above, just an explanatory illustration):

The left panel presents a typical scenario of binding experiment: signals Y are measured in two experiments over the same set of X values. It results in pairs of points with the same X-s and two diferent Y-s lying one under another on this coordinate plane.

Two other panels are Scatchard plots: the plot Y vs. X is redrawn on an extremely awkward coordinate plane Y/X vs. Y

Central plane shows correct appearance of Scatchard transformation of data on the left panel. Both horizontal (Y) and vertical (Y/X) coordinates on this plot are linear functions of measured signal Y. So, now these pairs lie on the lines drawn through the start of coordinates.

Right panel presents fabricated Scatchard plot: pairs of datapoints can not lie precisely one under another on this coordinate plane.

Obviously, if you try to fabricate Scatchard plot, you would not first fabricate the picture at the left and then carefully transform it to a trustworthy picture at the centre. Almost certainly you would immediately fabricate the Scatchard plot. Then, most likely, you would draw a picture like that on the right panel which perfectly fits the common notion of a usual "nice looking curve". This type of folly is rather widespread; I've even made an attempt to develop some statistics for quantitative measuring sci. fraud on that base (see Statistics of Sci.Fraud).