Yes. Yes I can. There are regions of values in Park-Miller that are "toxic". The values are far from random. Without reseeding, the generator tends to stay in those regions for short periods; usually about half a dozen values. Incrementing by 1 guarantees the generator will stay in those regions for hundreds of values.
The correct choice is to increment by a large prime number excluding seeds that are outside the Park-Miller range.
I believe this thread includes those details (link also included in post #4).
So, if I understand you correctly, this would mean that in this case the start of the pseudorandom sequence is more predictable between consecutive boots whenever the seed resides in one of these regions.
Sort of. "Predictable" isn't really a measure because the values for all PRNG are predictable.
Dice rolls are a good example. That's even what surfaced the problem. Imagine a dice gadget that's supposed to replace physical dice. Also imagine that for the first 1356 rolls the result is either 4 / 4 or 1 / 6. While extremely rare, that is possible with physical dice. Our gadget could actually be working correctly.
But it isn't. The random number generator is lingering in a toxic region because we keep reseeding with sequential values (1, 2, 3, 4, etc.). The low bits, which are used to determine the result, don't change much if at all in the toxic regions.
There is a second problem with incrementing by 1. Half the possible values (zero and everything greater than 2^31-1?) for Park-Miller are essentially toxic. I think they all fold to zero or four. However, it will take a great while to reach that region when incrementing by 1. But, once there, it will take equally long to get past it.