I stand corrected, there are in fact a decent number of people who mentioned HMAC(K,X)==HMAC(H(K),X) when len(K)>blocksize. So, you had no reason to cite Solar Designer and I shouldn't have implied you should. Sorry about that!

Still interested if you can find an earlier cite for HMAC collision. Most of what I find implies mad crypto cleverness, not this silly little construction.