gevent has some serious drawbacks. Stack traces are incomprehensible, it works by monkey patching existing code, and it falls apart if you have a blocking operation gevent can't make async. Loom is a bit like monkey patching, but at the VM level, so I expect it to be much more stable.

It's worth mentioning those drawbacks are implementation problems, not soundness or ergonomic problems with the model itself.

I would consider myself a pretty harsh critic of Python but even I appreciate the elegance of the gevent approach to concurrency.