I like the idea, and nearly everything you can think of is possible if you do it as a rather informal fun event where the participants can trust each other to play by the rules.
The problems start as soon as this is not the case and there is no way to prove beyond doubt that everyone played by the rules. In particular, it is trivial to have the BOINC client report a wrong number of cores, not too hard to fake the reported CPU, and possible to download and report tasks on one computer but crunch them on different computers.
On a different note, AMD's mainstream desktop CPUs currently have up to 16 cores, Intel's high-end desktop CPUs, let alone workstation and server CPUs, can have even more cores, so there might already be several potential participants who exceed the 8-thread limit with all their competitive CPUs (I'm planning to join that group soon). A better idea than a limited number of cores might be to normalize the time by number of cores.