This original linear power supply circuit used in most pinball machines was designed by Bally in 1976. It's the same circuit that pretty much everyone used in their linear power supply through the 1990s. Note Bally added a resistor (r50) in this circuit to slightly increase the output of the LM323 voltage regulator. It's how the LM323 works - it is an adjustable voltage regulator, as shown on the LM323 data sheet. For some reason Williams never implement the adjustment resistor in any of their power supply designs (until WPC95, as Jwart stated.)
Note if you check the +5 volts with a DMM on any 1977-1985 Bally -17/-35 games, the +5 volts usually measures 5.05 to 5.25 volts. This is because of this r50 resistor. Compare this to the +5 volt circuit on 1977-1985 Williams system3 to system7, and you'll see 4.85 to 5.05 volts (the average is around 4.93 in my experience.) So the lack of this resistor in Williams' power supply circuits gives a slightly lower +5 volt output from the LM323.
So does this matter? Well on Williams system3 to system11, no, it really does not matter, the CPU will usually run fine at 4.85 volts. But with Williams WPC, they implemented a Dallas reset circuit into their CPU board. This is where problems happen. The Dallas circuit monitors the +5 volts, and if it see it dip below 4.8, the Dallas chip forces a CPU reset (game reboots.)
So why the dip in power? Well when you pulse the 50 volt coils using the flippers, this puts a lot of strain on the transformer. Unfortunately, this causes ALL the voltages to dip slightly. (You can see this on a normal game when the General Illumination lights dims slightly with an intial flipper "flip.") When this happens to the unregulated 12 volt AC line, on a normal game, the filter capacitor and voltage regulator makes up the difference, and keeps the +5 volts DC steady. But add time and age to the mix, things get old and don't work as well, and this can cause the +5 volts to dip too much, and the Dallas chip forces the CPU board to reset.
So, what's the fix? Well in a perfect world, replacing the +5/12 volt filter cap (C5 and C4 on WPC) should fix the problem. Unfortunately it doesn't always work. (There's other factors in the equation, which I won't go into at this point.)
The other issue is the LM323 itself. These are not perfect devices. That is, if you take ten brand new LM323 voltage regulators, put them in the same circuit, you won't always get 5.00 volts as the output. The spec sheet on them shows a variance of 4.8 to 5.2 volts (if i remember correctly.) Because of this, Bally implemented the r50 resistor. In case they had a bad batch of LM323 regulators, using the r50 resistor always ensured Bally got 5.00 to 5.25 volts. Bally err'ed on the side of "a little higher is better than a little lower." Williams unfortunately didn't do this, which is fine on system3 to system11 games. But when put in conjunction with the WPC Dallas reset chip, you sometimes get WPC game resets when using the flippers. (Especially on games with more than two flippers like say Twilight Zone, which strains the transformer more.)
So what's the solution? Always replace the original C4/C5 filter caps. You can also buy a new LM323 and hope for the best. Remember though, you could be replacing the original LM323 with a 4.95 volt "native" output with a new LM323 that only outputs 4.90 volts! (So a new LM323 isn't always the answer.) Unfortunately, add to this that LM323's aren't cheap, and they are now obsolete (meaning there's a limited supply on this earth.) Because of this, I feel that implementing an R50 style resistor into the WPC linear power supply circuit is a good fix for the problem. It will bump the LM323's output slightly higher, preventing the Dallas reset circuit from overtaking a game and causing endless resets with flipper action. Sometimes it's just the solution needed to fix a game without replacing a long list of (maybe obsolete) parts.