One of the major challenges in a Software as a Service (SaaS) cloud, is the fault-tolerant and cost-effective scheduling and execution of the end-user applications within strict time constraints, in order to provide results of acceptable quality. Towards this direction, we investigate the performance of strategies for the scheduling of fine-grained parallel applications in a SaaS cloud in the presence of transient software failures, which may occur during the execution of the applications. We combine and compare by simulation different techniques that incorporate application-directed checkpointing and approximate computations, in an attempt to: (a) provide resilience against temporary software failures, (b) guarantee that all applications will meet their deadline, (c) provide application results of as high quality as possible and (d) minimize the monetary cost charged to the end-users for the execution of their applications.