I'm sorta disagree at last point. 20 years stuff is already drifted most of it's stress out, so it is _usual_ to have older meters being more stable. Of course that does not apply to parts like capacitors (which degrade with age) or connectors/keys/displays, but real important stuff in 3458 and other hi-end meters usually properly guarded or designed in a way to minimize those drifts. Caps are easy to replace, so with bit of knowledge, it's possible to get good device, given all important stuff is good. One of latest service notes for 3458A actually does imply that newer meters had excessive drift issue, as newer parts may be not as tightly manufactured as older ones.
In any case you will learn a hell lot about these sensitive designs in case your meter go wank or need repair. Even with all shenanigans described above, I learned more in these 200 hours spent than few years messing with Keithley 2001's (which are meter of itself already to start with). This also goes far beyond repair itself, but also how to produce stable enough signals and references to test it, how to maintain proper wiring and connections, how change in ambient temperature affects parts and so on, it goes and goes. I don't regret on this project, even though it makes me angry half of time I touch it. It's worth those $$ spent.
Also I don't think it's fair idea to send self-repaired and bodged up units to Agilent to repair. As they cannot guarantee by any means that parts you changed by yourself will meet specs/reliability requirements they have for it, so it may just be as simple as replace all affected PCBA's, which would cost way more than calibration fee. That will also prevent you from going into unit again, which sorta kills lot of learning right here. So, no, I'll not send mine to official calibration, it's not worth that. Would be more interesting (and way more useful!) to build and calibrate 10V/10k references and make generator suitable for SCAL.