Having GPU miner as part of the release bundle at very launch has to be mandatory for coin to be considered as "fairly" launched. CPU2GPU ratio is irrelevant. Even if it is 1:1 it simply has to be released immediately.
I disagree. We are eagerly awaiting for a coin that would *never* have a GPU miner, only CPU. There are people working on this, and Cuckoo Cycle looks the most promising. First coin which achieves this may be a revolution, with millions of individual miners finally being able to get involved in cryptocurrency.
The botnets indeed eagerly await this. I'm pretty sure this is a philosophically wrong goal, but we can debate that another time.
(n.b. -- I think there are GREAT uses for functions that have rough speed parity across hardware, but it's more from the perspective of things like key derivation, which is what scrypt was originally intended as, and not mining.)
And let's wait to decide about Cuckoo Cycle. I haven't even started on the design of how I'd do it on the GPU -- I'm still trying to solidify some of the algorithmic issues behind the mining process before I try to take more of tromp's money. We're still doing some back and forth about the time/memory tradeoffs, and I don't think the improvements are done yet: https://bitcointalk.org/index.php?topic=707879.40
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.