HP also uses this method. I have the Spectre x360 13 tiger lake i7, and I also do note this cycling behavior, but it is at higher power limits and temperatures. In performance mode, the PL1 when temperatures are low is 29.5W, which is a bit higher than the XPS. When initiating a stress test, the temperatures skyrocket to 100, and sit there, while the CPU continues to draw 35-40W. Then, the turbo time limit is up, and it drops to 29.5W in 15 seconds after sitting at 100 degrees. At 29.5W, the temperatures are at around 85-90 degrees, and it sits there for a little bit, and then starts dropping, .5W every second, until it reaches 20W. At 20W, temperatures are around 70 degrees, and then after around 10 seconds or so at 20W, it starts increasing, once again at .5W per second, until reaching 29.5W. Then, the temperatures are a little lower, but still over the 80 degree mark.  It sits at 19.5 W for 15 seconds and then drops back down to 20W. During this time, the fans are running at max speed and do not fluctuate. In balanced mode, the PL1 is 18W and drops to 14W, with temperatures in the 60 degree range.
NOTE: When gaming, for my Spectre, the power limits do NOT cycle. It sits at 29.5W constantly (temperatures around 80) and does not decrease to 20W. The GPU clock is at max frequency all the time (1300 MHz) and CPU is at 4 GHz. The frame rates do not cycle, and stay consistent.
Honestly, I see HP's approach to dynamic tuning as much better than Dell's. Not only did they use a triple heat pipe design (two leading to fans and one for passive cooling), compared to one for the XPS, but they also kept power limits higher and when gaming, the power limits do NOT cycle, which is good for gamers.

The other potential issue with cycling is fan speed fluctuations, which is an issue in some reviews.

As for difficulty determining an ideal speed/temp point due to varying environmental factors, I would think they could use an algorithm that oscillates back and forth with reducing deviation while evaluating the temp at each speed to settle on an ideal, e.g. too hot at 3.5GHz, thermal headroom at 2.5, too hot at 3.3, room at 2.7, and so on until settling at a speed it determines as just right, then adjusting in the same way from there when the environment, external or internal, changes. Of course, that would take a little more effort and, even though it would pay off on every device, they probably don't want to make the initial investment.

Also, I agree OEMs should allow the user to (easily) switch dynamic mode on and off. Not only to eliminate fan and FPS fluctuations, but to allow for apples-to-apples comparisons.

Of course the Dell is faster. Because even the low of 2.5 GHz is more than what the Asus offers. Since the relationship between power (heat) and frequency isn't linear, I imagine optimum strategy for performance in sustained loads would be stable frequency that gives you the temperature you want at equilibrium (you can boost in the beginning, but then you should settle down). Efficiency lost in those bursts will never be recovered in the cool down. Running at stable, say, 75 °C should result in better scores with none of the fluctuation (who knows, maybe even 70 °C could give interesting results). The problem is that determining the correct frequency is not trivial as it depends on environment. Oscillating between two modes is much simpler.

The point you're making here is even more relevant than you realize: the 9310 2-in-1 has a totally different cooling system, resulting in big changes between what is even otherwise the same single laptop model.