Extrachromosomal DNA (ecDNA) has emerged as a key driver of oncogene amplification and a major contributor to rapid intra-tumour heterogeneity, thereby promoting tumour progression and therapeutic resistance. This heterogeneity arises from pronounced cell-to-cell variability in ecDNA copy number, enabling complex ecDNA amplicon compositions within individual tumour cells. Approximately one-third of ecDNA-positive tumours harbour multiple co-selected ecDNA species. However, the mechanisms governing the heterogeneity and persistence of ecDNA variants - beyond the presence of distinct ecDNA species - remain less well understood. In particular, little is known about the maintenance of genetic or phenotypic diversity within a single ecDNA species. Here, we develop computational models to investigate the dynamics that enable the stable maintenance of tumour cells carrying multiple ecDNA variants ("mixed cells"). We explore how variant switching contributes to the persistence of ecDNA diversity under varying fitness regimes. Our results demonstrate that both a positive fitness of ecDNA+ cells and variant switching are required to maintain mixed cell subpopulations, whereas direct co-selection of mixed cells is not necessary. Notably, the fraction of mixed cells peaks at intermediate switching rates across fitness landscapes, a pattern reflected in subpopulation structures, transition probabilities between pure and mixed ecDNA states, and single-cell Shannon diversity indices.