As I understand it, the author conducted the test on an unusually slow connection. With a roundtrip time of almost half a second. Even a typical consumer line i

As I understand it, the author conducted the test on an unusually slow connection. With a roundtrip time of almost half a second. Even a typical consumer line is multiple times faster. He states that the workload is "saving one event to the database". And then sees 0.42s for his old solution and 0.48s for the serverless solution. Even if the server waits for the event being written to disk before it answers the request - that should be a negligible time compared to the roundtrip time he sees. So the "15%" only hold for his very slow connection. If he repeated the same test on a faster connection, then it could turn out the serverless architecture is not only 15% slower but multiple times slower. If 0.40s of those "0.42s vs 0.48s" he sees are due to his slow connection, then on a fast line it would be 0.02s vs 0.08s. Then his conclusion would be "4x slower" instead of "15% slower".