I don't have or collect any specific usage/benchmark info, but in my own experience the number of compute cores on a GPU scales pretty well with the performance of tyFlow's GPU-accelerated components. It's not necessarily a perfectly linear correspondence, since there are many other factors at play...but considering how many pieces of software nowadays rely on GPU-acceleration for various things, it's never a bad idea to spring for a GPU that's as powerful as possible, if you can afford it.
And keep in mind not all of tyFlow is GPU accelerated. The cloth/bind solvers are, as well as many of the new terrain components, but there are many areas of tyFlow that rely on CPU multi-threading speed and have no GPU acceleration...so if you're sacrificing CPU speed in order to afford a better GPU, that might not give you the performance boost you expect. Try to raise the performance of your entire system as a whole rather than running an insane GPU with a slow CPU or vice versa.
And keep in mind not all of tyFlow is GPU accelerated. The cloth/bind solvers are, as well as many of the new terrain components, but there are many areas of tyFlow that rely on CPU multi-threading speed and have no GPU acceleration...so if you're sacrificing CPU speed in order to afford a better GPU, that might not give you the performance boost you expect. Try to raise the performance of your entire system as a whole rather than running an insane GPU with a slow CPU or vice versa.