Cancer is caused by the accumulation of a critical number of somatic mutations (drivers) that offer fitness advantage to tumor cells. Moderately deleterious passengers, suppressing cancer progression, and mini-drivers, mildly beneficial to tumors, can profoundly alter the cancer evolutionary landscape. This observation prompted us to develop a stochastic evolutionary model intended to probe the interplay of drivers, mini-drivers and deleterious passengers in tumor growth over a broad range of fitness values and mutation rates. Below a (small) threshold number of drivers tumor growth exhibits a plateau (dormancy) with large burst occurring when a driver achieves fixation, reminiscent of intermittency in dissipative dynamical systems. The predictions of the model, in particular the relationship between the average number of passenger mutations versus drivers in a tumor, is in accord with clinical data on several cancers. When deleterious drivers are included, we predict a non-monotonic growth of tumors as the mutation rate is varied with shrinkage and even reversal occurring at very large mutation rates. This surprising finding explains the paradoxical observation that high chromosomal instability (CIN) correlates with improved prognosis in a number of cancers compared with intermediate CIN.