If you set in preferences
On multiprocessor systems, use at most 50.00 % of the processors
then 4 tasks will run faster. (but not quite twice as fast.)
Or if you want to go into computer Bios and disable Multithreading it
will have the same effect. Half as many tasks faster.

try setting your boinc advance preference on multiprocessor use 87.50% 7 task run faster and more effictve at least they do on my i960 3.20

If each task is too slow for you, I would suggest running projects with faster run-times, like HCC.

Otherwise, you can turn off HT, which will run 4 tasks only instead of 8, but it will be faster per work unit.

An HCC task will complete in about 40 minutes on an overclocked i7-2600k or i5-2500k both without HT enabled.

The idea of using multiCores to cooperatively crunch for a single WU competes with the idea that WUs can be sized and coded to crunch in parallel across multiCores. That is, if, say 8 cores, are cooperatively run and took 1-hr to crunch a given WU, that WU can be re-sized so that 1-core would take 1-hr to crunch. The output is the same: 1hr-perCore-perDoneWu.

To code for cooperative cores (multiThreading) is difficult and does not scale as well with the number of cores. To code for parallel execution (multiProcessing) is a lot simpler, and scales well not only with the processor capabilities, but also with the number of cores. Put simply, multiProcessing beats multiThreading.

Exactly, and that is why HCC is the project that would give birth to the first GPU-aware app in WCG! Rather than look for ways to run 4 or 8 cores (for a task that used to take CPUcores*WUruntime) cooperatively, it is a lot better in many respects to size-down a WU and run them in the hundreds-of-cores of today's GPU -- all in parallel.
;