This goes back to the question of the "turnout model" used by polling agencies when reporting results. In this election, it is very clear, and pre-election polling bears out, that "conservative" and "moderate" voters are more motivated to vote than "liberal" voters. In a normal election, the pollsters would look at voter registration and past voting trends, and predict what percentages of each class of voter would be likely to vote.
Typically, Republicans votes in greater percentages relative to registration, compared to Democrats. That is why, as you observe, Republicans in California, who are at a registration disadvantage, do better when the turnout is lower. This reasoning is also why Davis and the ACLU tried so hard to get the election delayed until March, when the Democrat presidential primary election would have occurred.
But this recall is an anomaly. Certain groups of voters are highly motivated to vote against Davis. The total percentage of votes cast by "conservative" voters, regardless of party, will be higher than in a normal election.
I believe the indications are that whoever wins the replacement vote will achieve more than 35% of the total vote cast. Therefore, the statement you made that "there are not enough conservative votes to elect anyone" is much more likely to be wrong than right. In fact, if Schwarzenegger or McClintock wins, it will be because of the conservative affirmative votes cast, not in spite of them.
"Self-described conservatives" is terminology used in polling. It is not unlike "probable voter" or "likely voter," where the pollster takes for fact the data reported by the respondant.