I have written code like that in the past, and would consider doing so again in the future. When, it's appropriate, it can be hugely beneficial.
When I last did it, I was wrapping a C++ API that I needed to compare against another dataset for a merge. The C++ API (which, admittedly, I also wrote), didn't have equality or hash operators defined on the Python objects. So, I monkey-patched them in for the keys I needed. It was actually the most elegant solution I could come up with as I could then naturally use the objects in sets and easily get differences to update the appropriate dataset.
As an aside, when I wrote the python wrapper around my C++ API, I purposely didn't define equals and hash operators for the objects, despite having full information to do so on the natural keys, because I wanted the flexibility to do the monkey-patching and override how the objects were compared depending upon circumstance.
