Hard to give a pat answer when your post basically says nothing is wrong.
Obviously something is wrong. What you need, instead of a bunch of parts, is an evaluation of whats is happening. Since its not likely for compression to be intermittant, its likely to be fuel or ignition. My first step would be to evaluate the ignition on an ignition scope and while on the analyzer I would vary the mixture to see if mixture was a problem.
You said the lambda control system was working. If you are right the mixture is being held to an average of .5% CO (if measured by an exhaust gas analyzer ahead of the cat). By average I mean that the richest cylinders are held to that mixture. If you have cylinders that are recieving say 10% less fuel those cyliders will exhibit lean misfire.
The way to evaluate this simply is to press down on the plate of the airflow meter. This instantly increases the mixture; if the car instantly smooths it means that you have a lean misfire condition. Solving the problem isn't as easy. BTW if your lambda system is working a small mixture change (by placing a finger on the plate)will be almost instantly corrected by the system; so the idle improvement will also be momentary. If you do this in small increments you will eventually exceed the ability to compensate and a rich mixture can be maintained.
The answer from here is tough as the 450's were kinda rough idling cars anyway. The 380/500 cars with basically the same system K-Jet w/lambda do a much better job running lean because the injectors are mounted to a swirl chamber on top of a separate idle air manifold system. The premixing and idle control smoothed the engine out at the lowest mixtures. (cars were never meant to run that lean)
If mixture doesn't change the idle I would verify and monitor timing. It should be steady far enough advanced.
The key to finding the answer will be in indetifying whats different between times of good idle from the rest of the time. Simple huh (bg)?
Owner 24 bay BSC
Bosch Master, ASE master L1
26 years MB technician