My understanding from Lenz is that it is unlikely to be before 6.0 and maybe not even then.
I would say it depends on which functions you are using and how complex the geometries are as to whether this is safe for production or not. I am using WL#1326 in production and have a table with half a billion complex spatial objects in it. I have not had a single problem, but mostly all I do is MBR queries on this table.
I have other smaller tables where I have attempted more complex queries which have crashed MySQL see http://bugs.mysql.com/bug.php?id=47429
The query in question was an intersects query where the input polygon was Scotland, a very complex polygon with 2812 separate rings and the data being cut out were addresses -- a point database with 29 million rows -- although the mbr of scotland reduces the actual intersection calculation part to 3 million or so.
This crashes MySQL hard in about 1 second. There are more efficient ways of doing this query, but that is not the point. Postgres, for what it is worth, and it pains me to say this, runs this query in a few hours.
I am desperate for MySQL to pick up the pace on spatial, but I fear it is a very low priority. Postgres is vastly superior in terms on spatial functionality and has much more active developement and even SQL server 2008 is now ahead. I have put a lot of time into building and testing these enhancements, which I don't resent for a second, but it would be nice to see a bit more movement.