You are right if you look at recall elections on a state level. The closest we come in recent California voting history was when there was a popular revolt against the antics of the Rose Bird California Supreme court in 1986. Although in that election, the the Supreme Court justices were essentially presented with a "vote of confidence" mandated by the California Constitution, three of the most liberal justices (Bird, Reynosa, and Grodin) failed to receive a majority vote and lost theire positions.
It is fair, however, to look at the dynamics of smaller recall elections (school board, county supervisors, etc.). Typically in those elections, an official or officials has run afoul of the law or some community sensibility. A small group of activists works to qualify the recall for the ballot. Once qualified, the activists generally turn out to vote in very high percentages. However, the activist percentage of the total vote is usually much smaller than that of the general voting population, particularly if the recall is not a special election (i.e. held in conjuction with a primary or other regularly scheduled election). The object(s) of the recall, also tend to rally their own activists and interests, but the vast majority of the voters end up in the "so what?" column and usually don't support the recall. that's why most recalls either fail to get onto the ballot or fail once on the ballot.
I think you would agree that this recall has caught the popular imagination. But I think the mechanics of the turnout are going to follow those of a smaller recall, in that the activists are motivated - but there are many, many more of them. Also, the "so what?" population is greatly diminished because most people realize that this election will effect them directly in the pocketbook. It doesn't matter what poll you might be looking at, the trends in the past week must be seen as bad for Davis and Bustamante in particular, disheartening for McClintock, and good for Schwarzenegger.
In years past, I have worked on a few campaigns and been involved with "what happened" analysis. This is where I have developed a healthy skepticism concerning polling data. Rarely does the polling input model accurately reflect the parameters of the election day voter profile. Complicating this is the increasing reliance on absentee voting. In 1992, when Bruce Herschensohn was running against Barbara Boxer, he was trailing by more than 20 percentage points in mid-September. The top of the ticket (George H.W. Bush) was doing poorly against the eventual winner Clinton, and Dianne Feinstein was trouncing incumbent John Seymour in the other, special election, for US Senate. But we were able to close hard on Boxer, by about 4% per week, and were within the statistical margin of error by the week before the election. Several polls confirmed this, putting Bruce down between 2-%; internal polls showed the race to be a dead heat. Then, on the Thursday before the election, Democrat leader Bob Mulholland conspired with the LA Times and other liberal media and dropped the "strip club" bombshell, which derailed Herschensohn's momentum and eroded into his overall support. As it was, Bruce lost on election day by about 7%, but won the absentee tally by an equal margin - those votes had already been cast before the dirty trick.
The post election analysis showed that Bruce had been badly hurt by the poor Republican election day turnout and the motivated Democrat turnout. Also, women turned out in record numbers and sided with the Democrats (particularly the two female senate candidates). The was "The Year of the Woman" - remember? Boxer did poorer with women than did Feinstein or Clinton, but well enough to win. The dirty trick did figure in the exit polling data, with women much more than with men. And, as I recall, Bruce did well with voters who were focused on "issues" rather than on gender or party identification. If the 1992 election had been held 1 week earlier, I believe Herscensohn would have won.
In 1992, only 37% of the voters identified as Republicans in the California contests. But only 41% identified as Democrats. The relative differences today have not changed greatly. This shows that it is possible for a conservative candidate to win in California.
I found some of the post-mortem statistical data. Keep in mind, this was for the 1992 California electorate:
Party Identification
Democrat 41%
Republican 37%
Independent and 3rd party 22%
Political Ideology
Conservative 29%
Moderate 49%
Liberal 22%
I wonder if I can find the 2000 demographics on the internet.