Plants interact closely with microbes, which are partly responsible for plant growth, health, and adaptation to stressful environments. Engineering the plant-associated microbiome could improve plant survival and performance in stressful environments such as contaminated soils. Here, willow cuttings were planted into highly petroleum-contaminated soils that had been gamma-irradiated and subjected to one of four treatments: inoculation with rhizosphere soil from a willow that grew well (LA) or sub-optimally (SM) in highly contaminated soils or with bulk soil in which the planted willow had died (DE) or no inoculation (CO). Samples were taken from the starting inoculum, at the beginning of the experiment (T0) and after 100 days of growth (TF). Short hypervariable regions of archaeal/bacterial 16S rRNA genes and the fungal ITS region were amplified from soil DNA extracts and sequenced on the Illumina MiSeq. Willow growth was monitored throughout the experiment, and plant biomass was measured at TF. CO willows were significantly smaller throughout the experiment, while DE willows were the largest at TF. Microbiomes of different treatments were divergent at T0, but for most samples, had converged on highly similar communities by TF. Willow biomass was more strongly linked to overall microbial community structure at T0 than to microbial community structure at TF, and the relative abundance of many genera at T0 was significantly correlated to final willow root and shoot biomass. Although microbial communities had mostly converged at TF, lasting differences in willow growth were observed, probably linked to differences in T0 microbial communities.