I'd like to add to PRRJIM's experience with 54" minimum radius. For decades I ran Westside J1 2-10-4's, Q2 4-4-6-4, M1's, 4-8-2, USH L1's Overland M1b, and virtually all the Sunset PRR power from a S1 6-4-4-6 on down around a railroad with main line super-elevated 54" curves (branch line 46" curves). None of the large Sunset locomotives required alteration. The Westside/KTM's J1's and Q2 did require some relatively minor mods to provide for additional lateral motion in drivers 1 and 5, and swing clearance for the rear trailing truck wheel. Once painted none of the changes were noticeable - even at eye level. My DC area O scale friends running heavy PRR, WM. B&O, and N&W rosters also use less than 72" radius curves to maximize operation interest – some in relatively modest basement spaces.
For appearance reasons you may want to go 72" radius and up - but don't feel you have to do it to run a large variety of O scale locomotives and full length passenger cars. My close coupled diaphragm equipped 80' passenger cars did just fine on 54"r. Wider radius curves do look better than tighter ones but they typically come at a price of track possibilities and aisle width - especially if the route is folded down the middle of the room. My recommendation is to go with a mix of radius. Wider in places where you will be viewing the trains from the outside of a curve, and tighter where the curved is viewed from the inside, or where it can be disguised by scenery features or structures (like a roundhouse complex). Building the railroad at higher elevations serves to improve the look of the railroad as a downward viewing angles tends to call attention to sharp curves.
While at it we should also mention track centers. The PRR minimum was only 13' (3 1/4") on tangents. My compromise was 3 1/2" on tangents and 3 7/8" centers on my one 60" radius curve. Changes to track spacing can nicely be made in the length of the easements in and out of the curves. The 3 7/8" centers will nicely clear 80' passenger cars passing my largest locomotives. If you plan to run Big Boys or other large articulateds with wide swinging boilers empirical testing can be used to find out what minimum track center they will need.
My recommendation is to size the minimum radius to the prototype equipment you want to run, and then use the layout design concepts in John Armstrong's book to maximize the operational interest of your railroad. A blend of learning about the prototype, buying/building/detailing models related to the prototype, and operating the railroad with friends has served to keep the hobby fresh for me.
Ed Rappe