Barring the use of a web_concurrent_start/end() set where one transaction ends within the concurrent window and another starts, what suggest is not possible as code is processed sequentially. How else might this happen?
The above code example would be an example of poor script construction which lead to a description of one transaction starting before the other ends.
Getting back to your core issue, statement N must complete (or timeout) before statement N+1 is processed. The only time this is adjusted is with web_concurrent_*() behavior where multiple threads can be executing concurrently.