Whip something up.
Seems to work.
Get it out there.
Hype the speed.
Get users to try stuff.
Get users who don't know any better to be wowed.
Get complaints from knowledgeable users who are familier with other systems.
Try to steal features from other systems.
Realize that the new features are incompatible with the old features, and decide to partially implement the new features in an attempt not to break the old features.
Fuck it up.
Break the old while not providing the new.
Scramble.
Pull out the new, attempt to put the old back in.
Declare the old baseline as the standard, and ignore the rest as long as possible.
Repeat the cycle.
Remember, MySQL is a commercial entity 1st, and is market driven.
Postgres is more of a well designed academic exercise that has spawned several commercial spin-offs. Things get put in because they are good to have, not because that satisfy a marketing need. So it is developed in a steady non-rushed fashion. Features "feel" right. It always has very high-end capabilities (log based hot-backup, moment in time queries satisfied against the log, not the changing data, etc) that blew away MySQL, and the SQL dialect done not end up surprising Oracle users, since they seems to have developed almost in tandem.
And now there is a vendor that took that Oracle aspect a bit further, and is attempting an almost complete emulation, ie: [link|http://www.enterprisedb.com/|http://www.enterprisedb.com/].
I'd suggest taking a look at it if you are an Oracle/Postgres fan.