A simple mathematical model (Granich et al., Lancet 373:48-57, 2009) suggested recently that annual HIV testing of the population, with all detected HIV(+) individuals immediately treated with antiretrovirals, could lead to the long-term decline of HIV in South Africa and could save millions of lives in the next few years. However, the model suggested that the long-term decline of HIV could not be achieved with less frequent HIV testing. Many observers argued that an annual testing rate was very difficult in practice. Small scale trials are nevertheless in preparation. In this paper, we use a more realistic age-structured model, which suggests that the recent high levels of reported condom use could already lead to a long-term decline of HIV in South Africa. The model therefore suggests that trials with for example 20% of the population tested each year would also be interesting. They would have similar (though smaller) advantages in terms of reduction of mortality and incidence, would be much easier to generalize to larger populations, and would not lead to long term persistence of HIV. Our model simulations also suggest that the age distribution of incidence has changed considerably over the past 20 years in South Africa. This raises some concern about an assumption presently used in EPP/Spectrum, the software used by UNAIDS for its estimates.