On second reading, it seems that you may be arguing here that there are some weak clients who take enormous amounts of time to process jobs and some strong clients (Quad-core systems, etc.) that process jobs quickly, and that the weak clients will be in trouble if the average job length goes from 8 hours to 24 hours. Economically, the value of a computer purchased five years ago is approximately zero. The contribution of a five-year-old computer to WCG, even if it ran 24/7, would also be close to zero. I would indeed argue that some potential clients are not productive enough to be WCG participants. Nobody should expect to be able to bring a Radio Shack TRS-80 or some similar museum piece to the WCG effort in anticipation of acceptance. If this is indeed your argument, then please specify an example of how old and weak a computer it is that you would argue WCG should be trying to accomodate as a client.

papa3: I think that there are very many 5-year-old computers out there on people's desktops, and WCG's aim is to include as many contributors as possible, so that the slower machines can make a useful contribution just by sheer numbers. If and when these contributors upgrade to something faster, hopefully they'll continue to contribute. If WCG were to eliminate their machines now, they probably wouldn't be around with their faster machines in the future.
Five-year-old machines can do most things that most people want to do, anyway - surf the Net, office apps, edit photos, play standard-definition video, recode video (slowly), run Win XP. HD video and modern games are out, tho. My 6-yo AMD Athlon XP2600 now sits on a friend's desk and does everything he needs it to do. The machine used by Astrolab, the starter of this thread, is probably in this category.
The older machines use more electricity per point crunched, but the environmental footprint of frequently upgrading computers is also huge. As well as manufacturing the machine itself, the chain of transporting, distributing, retailing, installing plus disposal of the old gear all need to be included.

As for BOINC WU crunch-times, I reason that these need to be small compared to the return-time allowance. Otherwise, not only will slow crunchers be unable to complete 1 WU in time, but BOINC will have trouble with job-queue management in general. Crunchers with intermittent internet connections would sometimes exceed deadlines unless they set their job-queue short enough that they would sometimes run out of work between connection periods.
It would be very useful if the amount of crunching needed for each individual WU could be estimated in advance. Then, smaller WUs could be diverted to slower devices, and if these data were also sent to our BOINC clients, they could calculate more accurate estimates of WU completion times, which would help with queue management.
However, I don't know whether it is actually possible for the scientists and WCG to predict the WU size. In the case of FAAH, this might be particularly difficult for molecules (ligands) that have not been docked previously. I guess that ligand molecular weight, or better, atom count, might be one predictor. ((Stats/regression analysis, Scripps/Sekerob?)). Then there are other projects to consider ...
[Edits: atom count, clarity of meaning]

Do to the troubles from this summer - especially given that forecasting actual duration is very challenging, we modified the process for FightAIDS@Home so that we send a very small number of workunits out for each batch earlier so we can get a rough estimate of the actual duration.

This has improved our ability to size the workunits and reduced the risks of major workunit duration issues like we saw earlier this year.

I have a broad range of systems ranging from a Sempron and Celeron up to a few pretty quick Athlon X2 64s. My concern was machine monitoring where a Celeron gets a FA@H job that requires 25 hours of crunch time and the machine is running at 70% and the computer is not on all the time. Its a pain to have to monitor this computer for 3 days to see if there is a problem with a single job.

My question was really, although I did not say it, can the system not allocate long WU to slow computers (especially when they are not 24x7), recognizing that the determination of what a long WU is and what a slow computer is are both arbitrary?

I am guessing that the answer is NO because a) The volume of long WU is low and b) as long as the WU is complete by the due date, WCG does not need to care how long the WU runs on one computer, which is an ok answer, I guess. That is, as long as the error rate on the long WU is very low.

Thanks Sek. I'm done with this thread