Assigning jobs to parallel machines is a classic problem in manufacturing and computer science. In many manufacturing environments, machines availability and job orders might change dynamically. The aim of reactive scheduling is mainly to repair a previous schedule. Reactive scheduling requires fast methods and knowledge rules in response to unexpected events. This paper presents some analytical rules on the dominance relationships on job mixings under Poisson job arrival and a method for estimating machine workload when there are alternative machines. It is found that uneven mixings of job types are better than even mixings in reducing setup time. An iterative procedure is shown to converge in workload balancing and the resultant makespan. Workload can be estimated with accuracy without running time-consuming optimization programs. Applications are demonstrated with numerical examples.