Statistical models of high-energy (typically above 10 MeV) solar proton fluences are needed to assess a broad range of cumulative effects observed in the solar system. Most models rely on a description of event fluences and occurrence based on a compound Poisson process for which the parameters are derived from observations. In this paper the effect of the size of the observation data set on the prediction of cumulated fluence by such models is assessed. The study shows that with the current size of the event data set of Solar Proton Events (about a couple of hundred events) the estimated parameters used as input to the model may lead to significant errors on the prediction of the quantiles of the fluence probability distribution function. Furthermore, it is shown that no significant accuracy improvement is expected unless the data set contains a much larger number of events (typically one thousand), which requires a long-term data acquisition infrastructure.