Having k= instead of api_key= in the query string, or limiting the actual api_key to X characters instead of 32 characters (md5 hash for example).
The speed difference would be so incredibly tiny -- six characters versus one character read -- that it is entirely inconsequential.
Keep in mind that while URLs don't have an RFC-spec length limit, keeping them under 2000 characters is general practice due to IE issues. Turning two thousand characters worth of query string into workable parameters is trivial, no matter what programming language you're using.
If you are seriously worrying about the performance characteristics of URL parsing, I'm going to wager that you probably need to sit down with a code profiler and find out your performance metrics actually are, because you're totally looking in the wrong place.