Phase I dose-finding trials typically are conducted using adaptive rules that select dose levels for successive patient cohorts based on the outcomes of patients treated previously in the trial. When patient outcome cannot be observed immediately after treatment, the problem arises of how to deal with new patients while waiting to observe the current patient cohort's outcomes. We consider two alternative approaches to this problem in the context of a phase I trial conducted using the continual reassessment method. With the first approach, a patient requiring treatment before the next cohort opens is treated off protocol with standard therapy, and otherwise waits until the next cohort opens. The second approach treats each patient immediately upon arrival at the dose recommended based on currently available data. We compare these two approaches by simulation under varying dose--toxicity curves, accrual rates, cohort sizes and early stopping rules. We evaluate patient waiting time, trial duration, number of patients treated off protocol and the probabilities of toxicity and of selecting the correct dose. We also study three strategies for assigning patients to trials when two or more phase I trials may be ongoing simultaneously. Based on our results, we provide practical guidelines for deciding among these approaches and strategies in a given clinical setting.