In reality, referees do trust that authors are making a good faith effort to present their results accurately (except for a little bit of cherrypicking and glossing over messy parts). As a referee, I'm simply not paid enough to search for fraud.
To search for mistakes, I skim a paper, think "yeah, that graph looks about right", [1] and give an opinion.
To search for fraud, I would need the entire source tree for the paper: data, programs, etc. I'll need to verify that the source code does what the paper says, and that the output of the program is really the source of the graphs. (This ignores the issue of fraudulent data, but that's really tricky to find.)
Since the data was destroyed (oops, I mean lost), this simply can't be done.
[1] I also check the proofs for mistakes, but that's irrelevant for this discussion.
Also I don't think this was the only data set in the entire world that relates to global warming. So it hardly invalidates all climate research if some random researcher formats his hard drive.
As for means, isn't the whole fossile fuel industry supposed to be behind the warming skeptics? They should have shitloads of money. In fact I wonder if by now they have their own research stations in the arctic, drilling for historic ice. I mean, they should? They really can not afford to buy their own thermometers?
I think they would probably have enough money to buy dozens of universities dedicated to climate research.
Climate science appears to have a lower set of standards.
Some of the CRU emails criticize climate journals which were considering adopting a "release all data" policy, and argued for not submitting papers to those journals. Other CRU emails discuss deleting data to prevent Steve McIntyre from getting it, and conspire not to release the data to him or other skeptics. This is why climategate is such a big deal.
As for other data sets, there are a few others. However, the CRU is one of the major sources of paleoclimate data. They aren't the only game in the world, but they are a big game. Both their data, and various analysis which are based on their data, are now inherently suspicious.
To make an analogy, imagine that a major open source regex library (maybe PCRE) is completely tainted (e.g., a rootkit). How large an impact would that have on open source software?
http://www.timesonline.co.uk/tol/news/environment/article693...
No, it's easier to assume trust, but science works best when there is almost zero trust in any single authority. This is why it's expected for you to present your methods, data, math, and then your conclusions. Ignoring malice a tiny mistake can drastically change an experiments outcome. So an experiment yet to be independently verified is next to worthless, but it's not sexy to work like that and it does not help your reputation.
Edit: IMO, what's missing from this discussion is the idea that independently verifying their results is a reasonable thing to do. This does not mean plugging their numbers into their code and looking at the results but starting from scratch using their methods and testing it yourself.
PS: One of the reasons that social science has such a poor reputation is the willingness to accept minuscule sample sizes as representing real research. A little less trust would go a long way to revitalizing those fields. Unfortunately neuroscience seems to be falling for this trap.