Part of the latency from memory read or write operations is for data to be input to or output from the data latches of the memory via an I/O bus. Methods and circuitry are present for improving performance in non-volatile memory devices by allowing the memory to perform some of these data caching and transfer operations in the background while the memory core is busy with a write operation. In the exemplary embodiment, when the multiple phases of a write operation vary as to the number of states to track, a phase-dependent coding enables efficient utilization of the available data latches, thereby allowing a maximum of surplus latches for background cache operations.