Background Unlike most virtual reality (VR) training programs that are targeted at homogenous populations, a set of VR games for rehabilitation purposes targeted at a heterogeneous group of users was developed. The VR games covered physical training, cognitive training (classification and reality orientation), community-living skills training, and relaxing scenery experiences. Special considerations for local older adults and people with disabilities were made in terms of hardware choice and software design. Objective This study aimed to evaluate the feasibility, acceptance, and efficacy of VR training among users with varying abilities. Methods A single-arm pretest-posttest evaluation study was conducted. The participants of the evaluation study were encouraged to undergo 30-minute VR training three times a week for 6 weeks. The 30-minute session consisted of 10 minutes of upper-limb motion games, 10 minutes of lower-limb motion games, and 10 minutes of cognitive games/community-living skills training/relaxing scenery experiences, as appropriate. On completion of each session, usage statistics were documented via the built-in VR software, whereas feedback on the experience of the VR games and adverse events was collected via self-reports and staff observations. Feasibility was reflected by usage statistics, and acceptance was reflected by positive feedback. In addition, health outcomes, including upper-limb dexterity, functional mobility, cognitive function, and happiness, were assessed at baseline, as well as 6 weeks and 3 months after baseline. The primary outcomes were upper-limb dexterity and acceptance of playing VR games. Results A total of 135 participants with a mean age of 62.7 years (SD 21.5) were recruited from May 2019 to January 2020, and 124 (91.9%) completed at least one follow-up. Additionally, 76.3% (103/135) of the participants could attend at least 70% of the proposed 18 sessions, and 72.5% (1382/1906) of the sessions had a training time of at least 20 minutes. Linear mixed effect models showed statistically significant effects in terms of upper-limb dexterity (small effect) and cognitive function (moderate effect). Among the 135 participants, 88 provided positive comments. Additionally, 10.4% (14/135) reported mild discomfort, such as dizziness, and none reported severe discomfort. Conclusions A set of VR training games for rehabilitation could be applied to users with heterogeneous abilities. Our VR games were acceptable to local older adults and those with different disabilities. Benefits in upper-limb dexterity and cognitive function were observed despite partial compliance to the training protocol. Service providers could refer to our experiences when developing VR training systems for their clients.