The best study which examines the relationship between sleep duration and mortality risk was published in 2007 by Hublin et al in the journal Sleep. Here's a link to the abstract:
http://www.ncbi.nlm.nih.gov/pubmed/17969458
It is part of the Finnish twin study, which followed over 20 000 twins over a 22 year period. This is an extremely large cohort, and the study had very high response rates. The analysis was thoughtful and comprehensive.
They showed that mortality rates were lowest for those who sleep between 7 and 8 hours per day. For those sleeping less than 7 hours per day, or more than 8, the mortality rates were about 20-25% higher. The results were adjusted for the covariates of education, marital status, age, working status, BMI, social class, drinking behavior, physical activity, smoking, and life satisfaction. Interestingly, and unexpectedly, sleep quality was not shown to be associated with differences in mortality risk.
The argument could be made that average sleep duration has a non-causal association with lower mortality. That is, people who happen to be healthier in the first place are more likely to have average sleep length. But another part of this analysis suggests that this is more than a non-causal association: subjects who changed their sleep duration during the course of this 22 year follow-up also changed their mortality rate, after controlling for the measured confounding factors. I suppose it could still be true that some other mortality-increasing factor was the cause of the sleep duration change, and not the other way around.
In conclusion, this data supports the commonly held belief that 7-8 hours of sleep per night is a desirable goal. It may be that particular individuals have a different "set point" for optimal sleep, and for those individuals optimal health might result from more or less hours than this average. Yet I do not actually see firm evidence of this in the research I've seen.
A 2010 meta-analysis supports the same conclusion: http://www.ncbi.nlm.nih.gov/pubmed/20469800 but I think the authors understate their findings. In particular, while a lot of the data showing increased mortality in short sleepers defined short sleep to be under 7 hours, the authors state in their discussion that "consistently sleeping 6 to 8 h per night may therefore be optimal for health." I think there is a significant difference between 6 and 7 hours, particularly due to pressures in the culture where many people are sleeping only 6 hours because of a busy schedule, while really needing 7 or 8.
Knutson in 2007 published a good article showing that sleep deprivation causes impairments in glucose tolerance (similar to the changes which occur in the development of type II diabetes), and impairments in the hormones associated with appetite regulation: http://www.ncbi.nlm.nih.gov/pubmed/185162
Here's one of the articles in the literature showing that sleep deprivation leads to an increase in proinflammatory cytokines and abnormal immune activation: http://www.ncbi.nlm.nih.gov/pubmed/19240794
I think it is especially true that if one has signs or symptoms related to sleep duration (e.g. feeling sleepy in the daytime after sleeping only 6 hours per night) then this could be taken as strong evidence that sleep duration should be increased up to the average (7-8 hours), if circumstances permit.
Patterns of sleeping long hours (above average) could be approached similarly, but of course if the reason for the long sleeping duration is medical illness or medication effects, etc. it would not be healthy to force oneself into a shorter (average) sleep regimen.
