According to FAQ post on BOINC: Estimated Tasks "Time To Complete" Are Totally Wrong by Sekerob

“The client benchmark is re-evaluated once every 5 days (140 hours wallclock)”

My benchmark run on 12.12.2008 is:

12-Apr-2009 00:26:23 [---] Running CPU benchmarks
12-Apr-2009 00:26:23 [---] Suspending computation - running CPU benchmarks
12-Apr-2009 00:26:55 [---] Benchmark results:
12-Apr-2009 00:26:55 [---] Number of CPUs: 2
12-Apr-2009 00:26:55 [---] 2881 floating point MIPS (Whetstone) per CPU
12-Apr-2009 00:26:55 [---] 6306 integer MIPS (Dhrystone) per CPU
12-Apr-2009 00:26:56 [---] Resuming computation

5 days later they are

17-Apr-2009 00:26:55 [---] Running CPU benchmarks
17-Apr-2009 00:26:55 [---] Suspending computation - running CPU benchmarks
17-Apr-2009 00:27:27 [---] Benchmark results:
17-Apr-2009 00:27:27 [---] Number of CPUs: 2
17-Apr-2009 00:27:27 [---] 2899 floating point MIPS (Whetstone) per CPU
17-Apr-2009 00:27:27 [---] 6275 integer MIPS (Dhrystone) per CPU

5 days later they are much lower:

22-Apr-2009 00:27:28 [---] Running CPU benchmarks
22-Apr-2009 00:27:28 [---] Suspending computation - running CPU benchmarks
22-Apr-2009 00:27:59 [---] Benchmark results:
22-Apr-2009 00:27:59 [---] Number of CPUs: 2
22-Apr-2009 00:27:59 [---] 2134 floating point MIPS (Whetstone) per CPU
22-Apr-2009 00:27:59 [---] 4954 integer MIPS (Dhrystone) per CPU
22-Apr-2009 00:28:00 [---] Resuming computation


In his article Sekerob does not go into any detail how the benchmark is executed or what external factors could play a part in the outcome.
What can be the most likely reason for theese low benchmark numbers? I have not changed any hardware during this period.