g_str_hash algorithm

["ANDREW MARLOW, BLOOMBERG/ LONDON OF" <AMARLOW1 bloomberg net>]

While measuring performance of my GTK app I noticed
that alot of time is spent in g_str_hash.
Google and a mail archive search tell me that this
routine has not been looked at for a few years.
I wonder what performance studies have been done.
I realise that there is a tradeoff between ensuring
the algorithm is fast and ensuring that few collisions result.
The more complex the algorithms tend to result in fewer
collisions but it takes longer to calculate.

I wonder if anyone has looked at the Pearson hash
as an alternative to what is currently done.

Regards,

Andrew Marlow