The most important thing to learn from Cuckoo Cycle is that he's doing it *right* -- he's treating his proposed proof-of-work in the same way professionals treat proposals for encryption or hash algorithms by publishing a clear spec and implementation, soliciting comments, and even funding (from his own pocket, no less) further evaluation and analysis. BEFORE it's used in a coin.
There is nothing inherently wrong with a coin as a testbed for a new algorithm. It is a case where the mining incentivizes something useful (testing and optimizing the algorithm), not unlike other coins where the mining process is intended to produce some useful scientific or mathematical result.
Where this ends up being bad is when the developer makes either dishonest or unsupported claims about the algorithm (e.g. "this won't be feasible to mine on a GPU") as opposed to being up front about the algorithm being experimental and unknown and seeing where it ends up.