Gene expression classifiers have been used to predict diagnosis of disease, patient prognosis and patient response to therapy. Although there have been remarkable successes in this area, a particular goal of personalized medicine is the ability predict a response from a single gene expression microarray. One aspect of this problem is the normalization of microarrays. Affymetrix GeneChip microarrays are typically processed using model-based algorithms that require all of the data in order to adequately estimate the model. We experiment with the RMA normalization procedure in an incremental fashion, adding new chips to an existing normalization model. Focusing on tissue-specific normalization models, we generate datasets that have very small differences from a batch normalization. Through several large datasets of patient samples, we provide evidence that RMA models of normalization converge to a common model in homogenous samples. These results offer the promise of maintaining large data warehouses of patient microarray samples without the requirement of constant renormalization.