I tried a similar project on two of my engines and the Railsounds ranged from completely erratic to nonsense. One electronic train friend of mine advised me that the engines that were supplied with Railsound tenders had capacitors on the motor, different resistance values in the motor windings, and some of them had shunted brushes. None of which my engines had.
Interesting... We had a steam Railsounds boxcar circa 1990 which worked pretty well with an unmodified postwar 2026 (obviously when pulling a medium-sized train.) Of course this was released in the days of one chuff per revolution, before traditional steam locomotives were really capable of scale speeds. In a related project, we added an Ott stored analog sound system to a different postwar non-whistle tender. That one did tend to begin chuffing before the train was moving. I think the Ott system had an optional connection to the E-unit which, for the sake of interchangeability with multiple locos, we chose not to install.
With a series-AC motor the speed is highly dependent on load. I would love to hear more about the motor differences Railsounds vs. non-Railsounds. It sounds like Lionel made an effort to tame this tendency and make speed more proportional to track voltage. Perhaps a similar technique could be used to improve the performance of postwar and MPC-era steam?
Most of the OEM railsounds installations I'm familiar with including the boxcar and the 2426W tender, use a hall effect sensor (magnetic wheel) on the tender axle to synchronize chuff to speed. I know the original poster didn't want to modify his tender, but it will be tough to add a sensor to the four-wheel Bettendorf trucks that came on the '78 Blue Comet. Good post!