Petra,
Your intuition is right on in regards to earthquakes of Ml>=3
in the San Francisco region.
Here are some statistics (I know how you LOVE statistics)

During the period from July 1901 to July 2001 (100 years), (a period
during which all felt earthquakes in San Francisco would probably
be recorded and thus a period where catalog completeness at about
Ml>=3 level is reasonably possible) there were 1224 earthquakes
of Ml>=3.0 in the area bounded by 38.2N - 37N, 123W - 121W.

Simple statistics demonstrate that the average time between
such events is therefore 36525/1224 = about 30 days. The numbers
of events in the past several years are:

YEAR # of Events
1989 166 (including 115 between Oct 19 and Oct 22 - Loma Prieta)
1990 34
1991 23
1992 17
1993 15
1994 15
1995 12
1996 10
1997 34 (including 12 on March 27)
1998 19
1999 8
2000 12
2001 9

Looking at these numbers, it becomes clear that many of these
events occur in swarms or aftershock sequences. For example, more
than 15% of the total number of such events can be described as
aftershocks of the Loma Prieta earthquake (around 187 aftershocks
of ML>=3 after normal background of about 12 events per year is
removed from 1989-1991 totals.
To get a truer idea of how often RANDOM events occur in the
SFB area we need to incorporate an statistical algorithm which
removes clustering, such as the Jones technique favored by Roger
and the EPR.
Using the Jones technique, we find that in the 2012 18-day periods
since July 1901, 596 have included at least one earthquake of
Ml>=3 in this region. Therefore the probability that, given
a randomly chosen 18-day period you would find a Ml>=3 event
in that period, is about 596/2012 = 0.29.
There would be 1 chance in 2 of encountering such an earthquake
within a 45-day interval. In the past 50 years, you would have
to choose a period of 130 or more days to be virtually certain
that you would encounter a Ml=3 in this region.