Partitioning a system among multiple input and output pin I O limited packages is a widely researched and hard to solve problem. We previously described a new approach yielding large improvements, which partitioned functions rather than structure, and which used a single bus for all inter-package data transfer. In this paper, we describe an extension permitting arbitrary distribution of I O among the packages, and highlight experiments demonstrating even further I O reductions as well as surprisingly improved performance, with nearly no penalty.