Version S (with the BF1 in it) had completely different physics, so of course S, T, U replays are incompatible with old versions. I am sure LFS won't even try to play them.
But S, T, U and U test patches are supposed to be compatible with all S, T, U versions.
There was one bug fixed in U19, where after the test patch with the new autocross marshall system (around U15 if I remember correctly) random numbers were affected and SP replays with wind enabled were no longer in sync. That was fixed in U19 so that S, T, U replays before U15 should now be in sync with U19 and U20.
I'm slightly concerned if any hotlaps are going out of sync... there shouldn't be any OOS hotlaps at all, though I can see that one posted in this thread goes out of sync some way into the replay. That could possibly be a one-off glitch during recording, similar to an overclocking / overheating issue so I can't draw any conclusions from a single replay going OOS. If there is a pattern of hotlaps being out of sync, I would like to know.
I don't know why the benchmark replay goes out of sync. Version T replays are supposed to be playable with the U versions. That is a bit of a concern, but I won't go investigating the replay OOS just because of that. There were some changes in suspension strength between S and U versions, though U is supposed to take account of that and play the replays correctly. I guess I should suggest that the benchmark replay is done again with the U version (even though I admit it's my bug).
I am more concerned if the hotlap replays go out of sync because there are hundreds of them, and I am trying quite hard to keep the versions exactly physically compatible in all these versions. But as I said, before I investigate any further, I really need to see a pattern or a means of reproducing it with 100% certainty. Meaning, a way to make a U replay go OOS in U20 or a U20 replay go OOS in U.