UPDATE:::::::::::::::

PROE's in its final form the Documentation post and the Library Source Code (post #1) have been updated.

I'm currently recommending 8 passes / 64 byte block which makes PROE about 33% faster than a plain AES implementation and about even with a fully optimized one (on a clocks / byte basis). But PROE scales much better and on a Core2 processor I expect clocks / byte to drop by 30-40%. Also the 8 passes is on the VERY safe side, since I don't have the statistics/number theory background to cryptanalyze PROE to a conclusive extent.

After having time to analyze and modify the algorithm for several months I beleive it's results are extremely secure. But feel free to download the documentation (its in thesis form, refer to chapter 2 for detailed descriptions of how PROE works) and analyze it yourself.

Also if your new to algorithm optimization at an architecture level there's a well put together section in chapter 1 of the documentation that covers a range of optimization tips.

Update:
On a Core2 system (as I predicted in my thesis) the clocks / byte speed of PROE improves by ~50% to 6.9 Clocks/Byte on a Core2 E6300 1.8GHz processor.