Do all broad band mixers work this way, that the conversion loss varies obviously

I am testing a broad band mixer circuit now, as is shown in the JPEG of the attachments. The mixer IC is sim-14+, from the mini-circuits, with low LO power level of +7 dBm. The conversion loss is about +8 dB according to the datasheet with +7dBm LO. I use this mixer for up conversion, in which the IF port and the RF port is exchanged for input and output ports. In other words, the RF port is used as the output ports, and the rf frequency is about 5.2GHz. The IF port is used as the input ports, and the input frequency is about from 100MHz to 2GHz. It comes to a conclusion that the LO freqency is from 5.3GHz to 7.2GHz. I found that when I wanted to achieve an optimal conversion loss, I had to add the LO power to about +25dBm. The test coaxes have been tested, and the loss caused by them is not very much, not more than 3dB. What is more surprising is that the optimal LO power level is varying with different input frequencies. Such as the 300MHz is about +25dBm and the 1GHz is about +20dBm. Certainly, all the optimal LO power is much larger than +7dBm.
What is wrong with the circuit? In the datasheet, the VSWR of the LO port is more than 3. However, I have made a single sim-14 mixer circuit, which can work normally using +7dBm LO power.
In the JPEG, the u1 is the sim-14, and the c1 is a coupled capacitor with a value of 10 pF, 0603 size. The circuit before the mixer all work normally. ps, I have made some different values of c1, which caused almost no difference.
The line between the pads of c1 is a DC power line of the amp after the mixer, which is not photoen in the JPEG.
I doubt the mixer is not soldered right and I made several same boards, none of which works normally with +7 dBm power level.

Here is the datasheet of sim-14+.

Using any mixer as an upconverter is a bit tricky. Usually one only wants one of the two sidebands and the other should be rejected; this calls for a band-pass filter at RF output. Good mixers generate harmonics, and by adjusting a suitable phase of them by "phasing" offers a reduction of conversion loss, often the flatness can be adjusted by phasing.

As you increased the LO input power high over the specified +7 dBm, it looks like you might have damaged one or more mixer diodes. Try a new mixer and give more care to the RF output filter; adjusting the phase of the lower LO harmonics should be done by using a variable-length coaxial (or waveguide) section between mixer RF port and the RF band-pass filter.

---------- Post added at 17:36 ---------- Previous post was at 17:35 ----------

For an extreme flatness of mixer conversion loss, you can sacrifice some of the loss by adding a 2...3 dB attenuator instead of the phase adjuster.

Seems that your mixer is dead. To check if is damaged, just check one of the port isolation (LO-RF or LO-IF) if corresponds to datasheet. Be sure that you are using the right LO power exactly at mixer pin connection.

What is the power level of your IF signal?
What conversion loss were you measuring with the recommended (+7 dBm) LO power?

The IF signal level is about -30dBm, certainly not more than -10dBm. If the LO level is +7 dBm, the conversion loss is about 30 dB. If I make the LO level be +20 dBm in the input port ( care that it is not in the LO input pin), the conversion loss is about +10 dB.

Looking at your layout there's a couple potential problem areas. First, the long power trace running parallel and very close to your long LO trace. The vias will help, but long parallel traces can couple. Second, the power trace running under C1. Third, your 0603 10 pf cap probably has SRF around 2-3 GHz, and probably has PRF (very high impedance) right in your LO range. It's very possible that you're reflecting or coupling off most of your LO power before it reaches the mixer.

It is a simple question that the handes soldering is of many questions. When it was soldered by an expert, the question disappeared. Thanks for all replies.