I just found a block on the testnet with the code optimizations posted a few pages back. This was on an Intel Atom getting about 46 PPS. For experimentation, I also changed the nMaxSieveSize to 250000 and still got a block. I'm not sure how this will influence things in the long run though. The lower sieve sizes do seem to increase PPS at the moment though.
Edit: found another on the testnet, I'm going to try this on the main network now.