Identifying a subset of taxa that maximizes Phylogenetic Diversity (PD) is a cornerstone of quantitative conservation planning. Traditionally, PD is defined over a phylogenetic tree in which leaves resemble present-day taxa and the branch lengths capture the estimated evolutionary distinctiveness. While PD maximization is computationally tractable on trees with unit costs, the problem becomes NP-hard when transitioning to phylogenetic networks or to budgeted versions in which protecting taxa incurs non-homogeneous costs. In this paper, we address these two challenges by providing definitions and a comprehensive analysis of three distinct variants of budgeted PD on networks. We conduct our study through the lens of a small structural parameter, node scanwidth (nsw), which measures the "tree-likeness" of a phylogenetic network. We show that two of the considered variants can be optimized in O*(2^nsw B^2) time, where B is the budget. For the computationally harder, third variant, we provide an algorithm to compute PD scores in O*(3^nsw) time. We further contribute the first exact algorithms to compute node scanwidth, recognizing that the utility of algorithms based on nsw depends on the ability to compute nsw and its corresponding decomposition. Our approaches integrate data reduction rules, dynamic programming, and an Integer Linear Programming formulation. We validate our theoretical results through extensive experiments on highly reticulated, simulated networks containing several hundred taxa, using heterogeneous costs. Our implementation computes PD scores and optimal nsw in fractions of a second, even on the most challenging instances. Furthermore, our budgeted optimization algorithms significantly outperform existing benchmarks for computing PD on networks, which were previously limited to unit-cost scenarios. The software makes analyses even on networks with a thousand taxa tracta...