That's because you really need an extra sound source for Diesels: the combustion noise (auto-ignition) radiating from the block. You're lucky if games model the intake and exhaust separately, externally at least, never mind something that's classically ignored (because it's usually relatively quiet in spark-ignition engines). In this case, you could get away with lumping it with the mechanical noise of the engine (like you can with spark ignition), but that mechanical sound only really changes with engine speed, and only slightly with the torque it's generating - whereas the combustion noise literally follows your right foot. So you really need four sources: mechanical, combustion, intake and exhaust.
It's not that difficult to record it all in action, but mapping all four sources (or more if you're being extra fancy and recording chassis-borne engine noise for internal views) in the right way in-game is a pain, especially if it's difficult to isolate the individual sources from each other in the recordings. That's because all four sources have different dependencies on "throttle" position and rpm.
The other way it can be done is to increase the variation in throttle position recordings; so instead of just on and off throttle recordings, which seem to suffice most of the time, you'd have to introduce several part-throttle recordings to blend between. That might only be necessary for interiors, but separate interior recordings requires more memory or less quality for a given system.
If you were modeling the engine (i.e. a "simulation"), the four sources can be generated separately and the controls can be physically modeled also. It's harder to do it that way, though, especially when it comes to "mixing" the resulting sources in a spatially believable way, but once that work's been done, you'd have a very powerful tool indeed.