Guys, I changed maxcoin.cpp to use two CUDA streams in an alternating and overlapping fashion,
like it's done in scrypt.cpp and scrypt-jane.cpp. This should get the same performance boost
as the previous commit, except that it doesn't raise the CPU utilization so much.

It's really hard for me to verify that this works though. If someone able to compile the sources
would report back if he can still pool-mine at the same Yay! rate as before, this would be great.
I don't have a pool setup for doing testing on my own, that's why I am asking...

Christian