I'd be willing to bet that the win percentage for both products will be very similar if the sample size is large enough. In fact, I'd bet they will be so close you won't be able to tell whether the difference can be attributed to the quality of the product or randomness. The ROI result will at least partly be a function of whose figures are contributing more money to the pools - also not a matter of quality.

I've done some smaller studies on different sets of speed figures that I thought were of high quality (never TG and RAG though) and they always produced very similar net results despite the fact that they often disagreed on who the fastest horse was. It was enough to convince me that very small differences in the figures don't matter much to the win percentage and that it's highly likely that no one has a monopoly on mistake free figures when we are talking about a length here or there.

I think the real trick is those occasional races where there is a huge discrepancy due to either methodology or interpretation. IMO, those don't show up often enough to infuence the aggregate results heavily, but if you have a preference for some reason, IMO that should probably decide whose figures you use.