Maintaing the simd improvements apart is a viable solution too, if you can separate them enough that they can be activated by a configuration setting they should not get into the way of normal operation.
hmm ... it should be possible to do that ... i mean ... i should be able to add a runtime flag to the runtime cpu detection ...
on the other hand, the code simd instructions, the vectoralignment and the threadlock for garrays it hard to split appart...
cheers ... tim