Doubts on mixer as reciprocal device (diode biasing in hybrid 90deg coupler)

I have doubts on mixer as reciprocal device.
Lets look at single balanced mixer with 90deg hybrid microstrip coupler. It has diode pair connected in series:
http://www.qsl.net/va3iul/Homebrew_R...wave_Mixer.gif
http://www.qsl.net/va3iul/Homebrew_R...brid_Mixer.gif
and rat-race mixer:
http://www.qsl.net/yo4hfu/Files/TG/R...96%20mixer.pdf
It works good for receiving (obtaining an IF frequency).

Although, let's try to use this mixer as a Transmit Mixer. Now we feeding this mixer with some kHz-range signal on IF port, connected directly to the center pin on mixing diodes. And providing some LO signal. Transmit mixing performance is not good (from experience).

Now if DC-blocking capacitor added in series with diode pair, and IF signal is fed at the point between capacitor and diodes, transmit performance improved. Why?
I modified the image a little to explain the idea (low pass filter part omitted):

I think the same applies to rat-race mixer too!
Also, can such diode pair+dc block capacitor configuration be used for receive mixer? Any drawbacks/advantages?
I think that it can improve voltage swing for RX mixer and less LO power can be used (more sensitive). Not sure.

If you add a DC blocking cap in series with one of the mixer diodes, the mixer will not work anymore as a single-balanced diode mixer. Would be a single-diode mixer, which is the simplest form of an RF mixer.
For reasonable up-conversion performances you have to use a double-balanced mixer, which provides better ports isolation than other mixer types.

Thank for you reply! I found this idea in double balanced upconverter which uses very low power LO source. Harmonic after microstrip filter, no amplificatiom. They put two capacitors, one for each diode pair. IF port is replaced with quarter wave stub (between. Diodes). The only difference between traditional mixer is that IF is fed not between diodes, but from one side of diode pair. Any ideas? Diodes are in series, but quarter wave open stub in center point makes short at RF.

I wonder if what you're seeing is due to how you're terminating the ports during the test. Depending on where DC blocks are used, the diodes may see differing bias and drive levels that may degrade performance.

In general a passive mixer should be reciprocal (when the LO is driven at the appropriate level). However if there is significant LO leakage to either port, then I feel that you may see some apparent non-reciprocal behavior. How exactly are you evaluating "performance?"