Specifically, when doing deep infodiving, it's not uncommon to juggle (that is, actually apply ops within 30-min intervals to) 80+ tabs; 2 visual studio in the background; local number crunching: large-ish data / machine learning sandboxes; one-click virtual-machines ("wonder how this looks like in a mac?"); and the occasional steam round. 2x4 vCPU cores usually humming along at ~10-40%, 20-something GB ram usage, no swap.
These reqs bias me towards performance, even at _some_ expense of mobility.