The work units are designed to exit after 8 hours of computational time. There are multiple calculations sent with each work unit. The first being the most important to the researchers. The others help narrow their search and are extra fine tuning. So the extra work done by the more powerful machines are not discarded, they are actually sent to the researchers. There will be more on this in the future when the project launches, but right now this is all the info we can provide.

-Uplinger

So there is a slow machine doing 8:00 hours being cut off and a more powerful machine doing all 16. As observed, the credit is a simple 1/2 of the total claimed by both. Would have expected a similar approach to HCMD2 or RICE... number of positions/seeds computed and proration, only when both doing equal jobs in a task-quorum the simple credit average being applied.

Not being reissued where there is no true quorum, yet used by the researchers may warrant credit consideration.

2 cents thought.

I had 1 linux machine that could not do the beta's, I did not have enough memory, the message said I needed 750 MB of memory and that was for 1 beta

I've been seeing credit proportional to number of "jobs" completed.
e.g.
BETA_ A.22.C17H12N2S2Si.2_ 1-- 613 Valid 1/06/10 03:57:29 1/06/10 15:23:45 7.37 132.8 / 86.2
BETA_ A.22.C17H12N2S2Si.2_ 0-- 613 Valid 1/06/10 02:29:41 1/06/10 12:33:08 8.00 34.7 / 75.4 <-- mine

Faster machine completed, while mine was terminated during Job #13 (i.e. the 14th job) and was credited with exactly 14/16 of the total of the faster machine.

It's a good approximation in the long run, but job duration varies widely, even within the one WU (one may take 20 times as long as another), so credit levels vary widely too.

Might you have an old BOINC version which has not been tuned to at least match the Windows claim level? By what's seen here your machine was treated as a severe lowlier, claim dismissed (is this frequent?), the highclaimer knocked down by whatever logic and then 14/16 taken when in fact really 13 were completed, 14 assumed. There are multiple algo's in place similar to what's seen for other sciences and as yet not worked out which when kicks in.