It may be worth considering recent i7 processors in this context, since they have the 'auto overclock' feature on workloads that are not very parallel... this may to some extent give you the best of both worlds.
(The way this works is, if there is not enough parallellism, the CPU will move all the work onto one or two of the available cores, put some/all of the other cores into a low-power state, and overclock the still-working cores a few speed steps).
Note that the earliers i7 processors did not step up that much, but I believe from memory that the latest ones do a fairly reasonable overclock. This means that mostly single-threaded games could increase the clock speed to run fast, and parallel workloads could reap the benefits of 4 cores, both at the same time.
(I'll see if I can find a link about this and add it)
It appears this technology is called "Turbo Mode" or "Dynamic Speed Technology" depending on where you read. Current i7s only overclock by up to 2 speed bins (typically 266MHz). However, the Lynnfield range when release will apparently be able to do 4 or 5 bins (533 - 677MHz), which would boost a 2.93GHz processor to 3.6GHz on non-parallel workloads.
Also note that this feature is apparently not dependent on how many cores are in use, but rather how heavily they are used. It is based on the amount of headroom in the TDP that is left so that the processor can overclock without breaking the thermal profile it is designed for. Note however that 'headroom' would usually be due to under-utilised cores, so the point still more or less stands.