From a medical perspective, it's highly plausible that starvation, cold exposure, torture etc could result in physiological changes in eggs/sperm that could be passed on to children. This is even more likely if injurious conditions persist during pregnancy - for example, a Holocaust survivor is still suffering from PTSD when she becomes pregnant post-war, so her hormonal status is different from someone without any traumatic experiences. This is entirely scientifically plausible and very different from the Lamarckian superstition where wearing heavy earrings causes your children to be born with big ears.
The research on mice and cherry blossoms is much more interesting, because it at least suggests a
possible mechanism for so-called "genetic memory". That said, it is a single study that hasn't been replicated (as far as I can tell). If you click on the
Nature link, even the original article has quite a bit of skepticism about the results.
Coming from a molecular biology background, I find it highly implausible to think that a sperm could carry enough any significant percentage of someone's life experiences. A human genome contains approx. 3 billion base pairs, the equivalent of 750MB of information. If you assume 10 bytes of epigenetic information for each byte of genetic information, you get 7.5GB of data in a sperm. That's one movie DVD's worth; nowhere near enough data to hold
all of the significant events in your life. It is plausible that hormonal+psychological conditions during pregnancy could impart a much larger amount of hereditary information, but this would be a strictly mother-to-child phenomenon.
***
Re: the Great Threadjack: Skin pigmentation is very much
genetically determined, not epigenetic. You can't flip someone's skin color just by exposing them to sunlight for a few generations - just ask any bloke from Australia, mate. There is a huge difference in skin tone between a really tan white guy and someone from India, Nigeria, etc. The phenomenon of "light skinned Africans", "dark skinned Frenchmen" etc are entirely due to interbreeding and not any sort of Lamarckian effect.
http://www.nature.com/jid/journal/v132/n3-2/full/jid2011358a.html