I am currently playing a lot with the new memcached interface to MySQL. Making MySQL a "NoSQL" solution. 
Why should I access the same data via SQL and noSQL protocol?  A simple noSQL protocol like memcached only has lightweight access methods like "set" or "get". No complex parsing, no optimizing, no execution of expressions. And this additional overhead in SQL can be tremendous: I did a set of SELECT statements only based on primary key. Buffer cache is 50% of the table size. With ENGINE=InnoDB it takes 7.6 seconds to read the data. If I switch to BLACKHOLE engine it takes 6.4 seconds! BLACKHOLE has no code in the storage engine. So queries on BLACKHOLE engine create only load on parser and optimizer but nothing in the storage engine. But if I run on InnoDB it adds only 1 second or 15% runtime. Obviously the main part of execution time is outside the storage engine. Erkan found the same behaviour …