Transactions commit on the secondary in the exact same order as the primary. If you were using read committed isolation, it would still suffer from gaps due to transactions completing in different lengths of time. The gaps may be more pronounced on the secondary due to latency, but definitely not due to the transaction isolation level being snapshot. That would only make sense if they were using nolock/read uncommitted on the primary.

The problem would exist on the primary as well in that case. It just wouldn't be as noticeable.