Depois de ler o que se foi escrevendo, cito, pela quarta ou quinta vez, este útil texto (.pdf):
One of the first issues that arises in comparing referendum and election campaigns is that of voter turnout. Evidence suggests that turnout can vary much more widely in referendums than it does in elections. In Switzerland, where referendums are commonplace events, turnout is generally well below 50%, and can sometimes be much lower (Kobach, 1993). It can however rise to considerably higher levels when a particular issue engages wide voter interest or when a more intense campaign is waged by interested groups. In U. S. state referendums, turnout is notoriously low, and can be subject to even more extreme fluctuations. Butler and Ranney (1994) found that turnout over a large number of referendum cases in various nations averaged fifteen percentage points lower than that found in general elections in the same countries. Cronin (1989) found a comparable rate of "drop-off" -- i.e. the difference between voting the candidate and propositions sections of the ballot -- in American state referendums.
However, there is no reason to believe that turnout in referendums is necessarily lower than that found in elections. The turnout in some of the more important European referendums has generally been comparable to that found in national elections (table 3), and turnout in the 1995 Quebec sovereignty referendum registered an astonishing 94%, higher than in any provincial or federal election. Other important referendums in which turnout registered higher than that of a comparable election (table 3) are the 1992 Canadian constitutional referendum (+5), the 1994 Norwegian EU membership referendum (+13), the 1993 Danish referendum on the Edinburgh agreement (+3), and the 1993 Russian referendums (+12, +5). But clearly, a referendum held separately on a less salient issue runs the risk of lower voter participation. The 1992 New Zealand referendum on electoral reform (-28), the Puerto Rico statehood plebiscites (-9, -12), the 1980 Swedish nuclear power referendum (-15), and the 1986 Spanish referendum on NATO (-11) are all cases in which turnout fell significantly below that of the most nearly comparable election."
Em resumo: pode ser alta, especialmente quando diz respeito a questões de "regime", mas é, em geral, mais baixa do que em eleições de "primeira ordem", como não podia deixar de ser.