It's said that running the same under 64bit os gains about 4-8 percent in speed. It can also happen that it runs slower in 64 bit, jaw down. Boinc is supposed to test this over a series of some 20+ tasks but cannot find if wcg has activated this feature, running a set with the 32 bit app, then a comparable set with 64 bit app. There were some user reports of the 64 bit being slower from the stable runtime hcc1 project. Presently with the highly variable runtimes we see it's hard to compare.

Requested in past, a reference workunit set, so we can ad nauseum test them on various platforms and configurations. A separate feed would be needed, one that will not go into my contributions . Of course, this will only serve a fraction of the volunteers. Most all only want the easiest form of idle time computing, else they zone out and disconnect. The auto-testing would take care of optimization, and with the fast database wcg has of the devices, cpus, osses etc, it really should be set and forget autonomously best performance assigning.

Whoops, Yes, you're right.