query profiling

I'm aware of log_statement_stats, but as far as profiling goes, it's a
bit crude.

Is there an easier way to determine, for example, all the queries that
take longer than 0.5 seconds? I'd like to do this on a production
system, so I'd like to avoid anything that would crush performance or
cause any kind of downtime.

I could probably make log_statement_stats work, I'm just looking for an
easier way.

Regards,
    Jeff Davis