Lexicon model 200

Not likely. If it has cheap carbon resistors replacing them with metal film might help a little, but most likely the problem is cheap chips.

Check with Jim W, audio upgrades.

Welcome to the wonderful world of quantization error.  While you might be able to tidy up the analog section a bit, the noise is a fundamental characteristic of those old boxes. The 200 and 224 were before my time at Lexicon (I joined right after the 480L shipped), so I never spent much time with them.  But I can tell you a few basics about the 480, and you can extrapolate the technology backwards another several years.

The 480 had an 18-bit word size (with a couple of extra bits that were used during math operations). Any time you completed a multiplication or addition and wrote results back to memory, you made a little more noise as a result of truncation.  The basic coefficient size was only 4-bits (you could do 8-bit multiplies but they took more CPU time).  So not only was the math inaccurate, you also created zipper noise when you changed the coefficient.  In the case of the 224 onwards, a characteristic of the verbs was that they were changing coefficients constantly--that was part of the randomization that all Lexicon high-end 'verbs had.  As long as you had some reasonable mid/high frequency stuff in the audio, you could kind of get away with it, since the source masked a lot of the noise.

By the time you got to the 960L, you get 8-bit coefficients and a 24-bit word size.  You still truncated, but at least the randomization noise was down another 24 dB.  Then in the PCM96, we had floating point.  The noise effects were pushed pretty much below audibility.

If you want to experience this in a really stark way, take one of those old Lex boxes and put a 20-30Hz size wave through it.  Then sit back and listen.  You hear a constant grinding sound.  That's pure quantization error right there.