It's quite frustrating to run for 450 hours and then get a computation error (this for hadcm3n...). Are the results still useful with this failure? I'm tempted to quit CPDN if I'm uselessly keeping other projects from running. But if the results are still useful with computation error, so be it. Thanks.

The purpose of this project is to find out what happens to a model when it's started with certain forcings (variables), each of which can have a large range of values, and then see what happens to it.
Some run for the full amount of time allocated, some don't. The ones that do complete are then started again from that point and let run to see if they fail in this new time span. Or the next. etc.
Eventually the point where they fail will be located, and the results of the starting values will then be known.

As for the hadcm3n being more prone to failures, this was announced in this thread in the Science section, which was pointed to by this post in our News and Announcements thread, to which every person crunching here should be subscribed, as it's where we post items that everyone needs to know about.