sql-benchmark | synthetic benchmark to compare the performance | SQL Database library

I have data with column zone and starttime. I want to query for n latest groups of data where group is defined as records with the same zones without other zones in between sorted by starttime.

In this example the n is 4. First group has 2 A's then 2 B's, then again single A and then 3 C's.

I have a query that performs this task correctly: http://sqlfiddle.com/#!17/ffbee/1 However this query is probably not efficient for large tables because it first selects all data and only later get's what it needs. I'm aware that this is probably possible to write using procedure but I wonder if I can achieve it using only sql in declarative way.

UPDATE

I have benchmarked original query, @Sentinel query and application solution that fetches 20 results at a time and checks if required number of groups has been achieved. The n is 4. The group sizes are randomly between 10 and 20. There are 4 zones. All solutions: Sentinel query and application solution:

Sentinel's query is optimal. It has constant complexity and is faster than app query. Thanks :)

Source code of the benchmark tool if anyone is interested: https://gitlab.com/virtual92/groups-of-data-timeline-sql-benchmark Graph source: https://plot.ly/~Vistritium/14/