I agree with you that trying to replicate someone's cherrypicked data is a frustrating experience. Also, I learned to never, ever accept material such as probes from anyone whom I wouldn't trust with my life. Much easier to spend the day or two and make it myself than to be lazy and use someone else's material just to discover after a month that it wasn't prepared properly.
I have to say one thing, though, and it's all of this really taught me to question "truths", to ask the "obvious" and "stupid" questions. Those are often very good questions and the answers can be very interesting, and sometimes there are no good answers.
Trust Noone.
My previous labmate was a cherry picker, and it drove me crazy! He got published in PNAS twice and Ange. Chem. for all of his "hard work." He was a favorite of our adviser, which only increased my resentment. And he didn't use error bars because there is no error in a single measurement.
So where is he now? He is in his post-doc, and will go out for academic jobs next year - only at the top 10 schools.
I agree that cherry-picking is worse than falsifying especially since others will probably be able to replicate the result by not reliably making it especially frustrating and time consuming.
I think there's a subset to #4 which is ok. This is when someone makes a mistake, but the mistake is written into their methods section. For example, they develop a new analysis method which contained a flaw that biased their results. In that case, the response could be another publication by the same person or others that corrects the flaw and advances the fields' knowledge. It's the mistakes of "we thought we did X, but actually did Y" which were bad.
Commenting by HaloScan
