The early boards with three tempcos and two 3046 tranny arrays (the most common VCO board) does drift.
But it can be minimized in many ways.
Experienced users usually come to understand one must allow warmup time, keep drafts away and not to expect absolute accuracy.
Five of the opamps on the board are ripe for replacement with lower drift types, there are lower drift resistors available, trimmer ranges can be limited, etc.
The range switching can be improved also.
The buffer board can have better opamps installed and the range switch contacts can be doubled-up for more reliability.
Very important is to make sure all summing resistors are closely matched too.
If they’re not, scaling the vcos can be more difficult and other inputs (the pitchwheel, for example) will throw the tuning off when used.
It’s come to an age where the original, closely-matched resistors may have drifted and so I stock complete sets of hand-matched precision resistors for my own restorations and for sale.
I have an IMF3954 here, but just one. (same part as 2N)
But there are other dual FETs that will work fine.
In the Odyssey and other Arps, they use 2N3958’s.
I had a bad 3958 once and no replacement, but did have a bunch of 2N5524’s (still do) and I found they worked better.
They are used in the keyboard circuit.
So you may find other dual FETs out there that might work fine, but just be slightly off in specs like pinch-off voltage or whatever.
Something minor that might not make much difference or even improve on the design in some way.
I should also mention that if you’re building any of these older VCO designs, be especially careful in your choice of integrator capacitors.
While it’s well known that polystyrenes are probably best, the most important spec to look for is a low dissipation factor.
There are some really nice caps, but lousy in this regard and if you use them, your reward will be lousy high frequency tracking.
If your DIY boards are like originals and you have long traces to these caps, you may also consider “flying” them off the board to remove the effects of board capacitance.
As to 901As and B’s, I’ve seen (and also built) my own version of the 901A.
In my opinion, some of the drift does come from the A, but most of the criticism I’ve heard is towards the B’s poor range switching.
Since it uses capacitor substitution, it’s almost impossible… even with matched caps… to get a pair or triplet of 901Bs to range switch together well.
One can employ basic R/C design conventions and add resistors to trim the octave settings.
These would usually be very high values in the megaohms if you’ve already matched your caps as close as possible.
The larger problem is that 901As and Bs aren’t especially linear.
They can track well over certain ranges, but their response isn’t usually a very straight line.
More like a slightly bumpy line.
So one can take two 901s, scale them the normal way by adjusting them back and forth at 1,2 or 3 octave intervals, but then find individual notes within those octaves imperfect.
It can be very frustrating.
On a “normal” VCO, if your widest notes are in tune, so are all the notes in between. Not so with 901s.
Both A’s and B’s can be temperature sensitive. The effect is especially pronounced though because not all modules will share the same drift characteristics and not drift the same amount or in the same period.
So one ends up with oscs that drive them mad because they all seem to have a mind of their own.
One funny thing: take a look at the Paia 4720 VCO schematic and one for the Moog 901B.
Both use a single unijunction transistor as their core and in the same basic way.
It’s amusing to see one of the most expensive oscillators sharing the same core design as one of the cheapest.