Since BOINC and SETIBOINC the 'time to completion' value shown has ALWAYS been WRONG !
On the same computer, same settings it has ranged from a few minutes to more than 20hours...
Recently it jumped from 3.48 to 6.12 hours whilst the system is dropping results (2 at a time with the HT) every three hours or so -even though I see variability in the time required to compute different WUs.
However, how is it that the CPU benchmark gives results that are so far off the reality ?? Could it correlate somehow, or store an avergae value of computing time and match its own results against it for error proofing ??

I assume that this is about the initial estimate? If so, the total Floating Poing operations and Integer operations are calculated from the times (one of the tasks Alpha does when there is a major change to the benchmarks) for a "typical" WU. These are then used with the benchmarks for your machine to estimate the time for a typical WU. Calculating the time for a particular WU involves actually crunching the WU at the server, and this would obviate the need for the DC community.

If the question is about the non-linear progress in S@H, that can apparently be explained by the fact that the early calculations take less time per iteration than the later calculations.