Is the 3% curved at all? If so I wouldn't be surprised if it was struggling with 20 cars given the dip in performance I've seen when curves were added. More testing will be needed but I'm thinking I may even need to alter my layout design slightly to have a tougher grade on the straight sections and a milder one in the helices.
You could spend a lifetime trying to work out an exact formula for what a loco can pull up any given grade & curve radius combination, but I think what I've tested so far gives me at least the name of the suburb that the ballpark is located in. If one F45 gives out with 29 cars on a 2% grade, then I think it's a fair assumption that-
-the same lone F45 would reach its limit with 14.5 cars on a 4% grade, or
-29 cars on a 4% grade if you used two F45s (assuming same curve radius, in my video it is 16" at the tightest point).
My current layout plan peaks at a 3.5% grade on the eastern helix that incorporates 11" radius curves (in short 180° segments). If the above assumptions are at least mildly accurate then the grade alone should be doable with two F45s + 29 cars, but I can't put a figure on what effect the tighter radius will have so who knows if the lower grade balances out the tighter curves.
Another thing to consider is exactly what constitutes satisfactory performance. Given the above video findings I could probably take the consist down to 25 cars and have the train clear the grade, but it would likely be grinding the rails pretty heavily towards the top. I don't mind a little bit of wheelspin (it's probably unavoidable), but if the train is noticably slowing down then I wouldn't be happy (I'm using TCS decoders with really good BEMF so resistance from gravity is mostly compensated for).
Given everything above I think that all three F45s could pull 29 cars up my 3.5% grade, but I'm not 100% confident they could do it without grinding the rails. On a positive note, my layout has a long bridge (split into two sections) that runs along the back of the layout between the two helices with the grades starting at either side. If I move the start of the eastern helix downhill grade to the middle of the bridge, the grade goes down to 3% (the western helix is already at 3.1%).
Lots of experimentation to do once I get a proper baseboard happening!