Modern variable data presses use substantial processing power. In many cases, a bank of processors is used to manage the RIP (raster image processing), and print jobs are performed using sophisticated parallel scheduling approaches. The high processing power of digital presses enables the possibility of performing valuable imaging tasks using the same processing units. Important imaging tasks include reading printed marks (such as barcodes), print validation and inspection.In order to optimize the interleaving of real-time printing and imaging tasks, different imaging approaches must be considered. In this paper, we consider three different classes of imaging optimization in order to compare their relative effect on throughput and on amenability to processing on the press. These are (1) performing down-sampling before image segmentation versus performing native resolution image segmentation, (2) selecting different programming languages/compilers (e.g. Java versus C++ in our experiments) for the imaging, and (3) marshaling images into a single buffer versus allowing the system to manage the image,. Our results demonstrate that, in general, changes in structural approach to imaging, such as (1) provides, have the greatest positive impact on processing, while (2) has the least impact. The impact of approach (3) is more highly dependent on the architecture of the press, and so is perhaps the method that can be most positively affected by intelligent modeling.