Here is my data for all the locos I have run. (C) denotes converted from 3 rail.
MTH ES 44 Diesel............................40" Radius
MTH L1 2-8-2 Mikado..............:......40" radius
MTH I1 2-10-0 Decapod...................44" radius (middle driver blind, per the prototype)
MTH H-10 2-8-0 Consolidation............40" radius
MTH GG1 (C)...................................36" radius
Lionel GG1 (C).................................36" radius
MTH Aerotrain (C).............................36" radius
MTH P5a (C).....................................44" radius (pilots fixed, this limits radius)
Sunset GG1......................................48" radius
Sunset M1b (C).................................44" radius
MTH Trainmaster.................................40" radius
Atlas Erie Built A-B-A.........................40" radius
Weaver BP-20.......................................44" radius
Williams K4 Streamlined......................44" radius
Atlas SD-35...........................................44" radius
Atlas RS-1.............................................36" radius
Atlas RSD-7...........................................40" radius
MTH F7 ABBA .....................................44" radius (after being reworked for closer coupling, before was 36" radius)
Sunset K4s..........................................44" radius
Westside PRR 2-10-4 J1a.......................44" Radius
That Westside was an outstanding scale model. It is a huge locomotive. It had been previously owned by Ed Rappe who reworked it to go around tighter radius curves. Primarily he added some side play in the drivers, but also trimmed back the insides of the cylinders I don't think even Ed envisioned it going around a 44" radius curve. To be fair, I did have easements to the curves. Nevertheless this was on a test track I built to evaluate for the real radius requirements for 2 rail. After watching that monster smoothly glide forwards and backwards through those curves, I came to two conclusions: 1) I am definitely going to 2 rail. 2) The need for humongous radius curves, from an operational viewpoint, is a total myth.