On-farm varietal diversity is now highest in marginal agricultural environments because, unlike the case in more favourable areas, modern varieties have not been a sufficiently attractive option for farmers to replace their landraces. However, the continued survival of landraces on farm is dependent on the continuing failure of plant breeding to provide better alternatives. Highly client-oriented breeding can produce suitable modern varieties for areas that were dominated by landraces. We examine here the case of the adoption of two upland rice varieties in three states in eastern India, Jharkhand, West Bengal and Orissa, where our surveys had confirmed low adoption of modern varieties in the upland ecosystem. The two new upland varieties were readily accepted by farmers who rapidly adopted them on large proportions of their land. On-farm diversity is maintained by the need to trade-off among varieties but once a variety with overall superiority was found this incentive was removed. The new varieties from the client-oriented breeding did not have weaknesses that farmers had to trade-off against their landraces so they were more likely to replace them. They also replaced older modern varieties. Farmers had previously maintained them along with landraces because they had to trade-off the higher yield of Kalinga III against the more stable yield of landraces or the higher yield of Vandana against the higher gain quality of the landraces. Only mean count per household could be used to test the significance of the differences between individual years. In all states, this revealed a significant reduction in landrace diversity with the adopting farmers when the addition of the two new varieties was not considered. Trends across years for mean varietal count, total count and Shannon–Wiener index showed a significant decline in Orissa, irrespective of whether the two new varieties were included in the analysis or not. In Jharkhand there was a significant decline only for mean count when the new varieties were not included. In Orissa, it was possible to test if rare landraces were as readily replaced as more common ones. Landrace replacement was unmitigated by increasing rarity yet the most rare landraces have the highest priority for genetic conservation. Some landraces were completely replaced by the farmers in the sample despite a diverse variety portfolio being a risk-reducing strategy in a region where there is a high risk of crop failure. There was little environmental heterogeneity in the marginal, drought-prone areas to slow this decline. The new varieties had an impact on the farming system because sometimes farmers brought additional upland into cultivation. They also introduced these varieties into more favourable, medium land. The strategy of releasing two new upland varieties met with some success in maintaining diversity.