I've started to add a pehash implementation to YARA. I decided to base my implementation on the description in the paper and only use the totalhash and viper implementations for comparing results. In doing so I've noticed some problems, and it is unclear who is right.
For starters let's take a look at running the pehash.py implementation from totalhash against a binary.
wxs@psh Desktop % shasum 4180ee367740c271e05b3637ee64619fb9fe7b1d2b28866e590e731b9f81de36