Here's an attention-getting paper from Tomas Hudlicky (and his co-author Martina Wernerova), and I'd like to help it get some more. It begins:
One who has been reading the literature concerned with organic synthesis in recent years, be it methodology, catalysis, or total synthesis of natural products, may have noticed considerable inflation in the values reported for isolated product yields, ratios of diastereomers, and enantiomeric excess values. A comparison of papers published during the period 1955 to 1980 with those published between 1980 and 2005 reveals that those from the more recent period frequently report isolated product yields of reactions >95%. Such large values were rarely found in the older literature and are all but absent in Organic Syntheses, a journal that only publishes procedures that have been independently reproduced. . .
There, does that sound like the chemical literature you know? Just a bit? Hudlicky has tackled this issue before, and the reasons he advances for the problem remain the same: pressure to make your methods stand out (to the scientific community, to the journal editors, to the granting agencies), a decrease in scale in reactions (making accuracy and precision more difficult), and, finally, what he refers to as "deliberate adjustment". That's well put; the rest of us know it as fraud.
He identifies the mid-1980s as roughly the period when things really started to go to pieces, saying that most procedures in reputable journals before that era are reproducible by, as they say, one skilled in the art, while the numbers have been decreasing since then. And he puts some numbers on the problem, performing a series of test experiments with extremely careful weighing and analysis.
These confirm what every working organic chemist knows: the more manipulations, the more sample you lose. Filtration through a plug of silica gel, into one flask, can give you pretty much complete recovery. But if you cut fractions, you're going to lose about 1%. And if you have to do a separation, even between two widely separated compounds on silica, you're going to lose about 2%. So people who report a >98% yield after chromatography from a real-world crude mixture are kidding themselves. The same goes for extractions and other common methods. In general, every manipulation of a reaction is going to cost you 1 to 2% of your material, even with careful technique. Hudlicky again:
Given that most academic groups do not subject day-to-day reactions to serious optimization or matrix-optimization  as is done in industry, it is reasonable to assume that the vast majority of the reactions reported in the literature do not proceed with quantitative conversions. Such aspect would approximate our experiments with mixtures of pure compounds. Because a minimum of three operations (extraction, filtration, and evaporation) is required in working up most reactions, we conclude that yields higher than ca. 94% obtained by work-up and chromatography of crude reaction mixtures are likely unrealistic and erroneous in nature. Such values may arise as a direct consequence of not following correct protocols, which would be expected in the fast-paced academic environment. (An astute student of the organic literature may discover that this very author has been guilty of reporting yields in this range from time to time!)
He goes on to detail the limits of error in weighing, which depend greatly on the amount of sample and the size of the flask. (The smaller the sample-to-container ratio, the worse things get, as you'd figure). And he turns to analyzing mixures of diastereomers by NMR, LC, and the like. As it turns out, NMR is an excellent way to determine these up to about a ratio of 95:5 , but past that, things get tricky. And "past that" is just where a lot of papers go these days, with a precision that is often completely spurious.
Here's the bottom line:
The conclusion drawn from this set of experiments points to the prevalence of serious discrepancies in the reporting of values for yields and ratios in the current literature. We have demonstrated that the facilities and equipment available in a typical academic laboratory are not adequate to support the accuracy of claims frequently made in the literature. . .The current practice of reporting unrealistically high isolated product yields and stereoisomer ratios creates serious problems in reproducibility and hence leads to diminished credibility of the authors.
He recommends a rigorous disclosure of the spread of product yields over multiple experiments, calibration of LC and GC apparatus, or (failing that) at least admitting that no such analysis has been done. (He also recommends getting rid of the concepts of diastereomeric and enantiomeric excess, in line with my fellow Arkansan Robert Gawley's advice). But I think that these ideas, while perfectly reasonable, don't get at the underlying problems - the inflationary pressure to produce more and more noteworthy results. Hudlicky's rules should be adopted - but I fear that they might just push the self-deception (and outright fraud) into newer territories.
I'm glad he's published this paper, though. Because everyone knows that this is a real problem - we complain about it, we joke about it, we mutter and we grit our teeth. But "officially", in the published literature, it's never mentioned. Let's stop pretending, shall we?